Research
Dissertation research
In my dissertation project I study the causes and consequences of anti-elite appeals in the context of electoral competition. On the one hand, I’m interested in what drives the adoption of anti-elite appeals: How does the interplay between demand side factors (e.g., public opinion or voters’ party images), supply side factors resulting from the strategic interaction among political parties, and contextual factors (e.g., political events or historical legacies) shape the incentives parties have to employ such appeals? On the other hand, I seek to explain how established parties respond to their challengers’ use of anti-elite strategies: How do parties being the target of the criticism implicit to such rhetoric respond to them?
Some of the more specific questions around which I currently wrap my head are:
- Why employ some parties anti-elite rhetoric as parts of their electoral strategies and others not?
- Why is it that particularly parties who are newcomers to a given political arena or positioned on its fringes use these appeals?
Measuring anti-elite rhetoric in textual data
The questions at the heart of my dissertation research are comparative and thus appropriate research designs require to study change in parties’ anti-elite rhetoric and electoral strategies over time. Yet, reliable, time-varying, and actor-specific measurements of anti-elite rhetoric are thus far lacking, and existing, expert-coding dependent measurement instruments are ill-equipped to yield the many data points necessary to implement such research designs.
In a joint project with Tarik Abou-Chadi (University of Zurich), Pablo Barberá (Facebook and University of Southern California) and Whitney Hua (University of Southern California), I work on developing a measurement instrument that enables measuring anti-elite rhetoric in large amounts of textual data. Our approach combines the strengths of a content-analytical measurement approach and crowd-sourced coding to obtain measurements of anti-elite rhetoric in for a large subsample of texts. We then apply natural language processing and statistical learning techniques to obtain a classifier that allows labeling uncoded texts.
The main advantages of this approach are is scalability, as we can obtained many measurements with relatively little human input. To facilitate the reliability of text-level measurements, we collect multiple codings per text from the crowd, and apply a selective expert review strategy to obtain gold-standard labels for a subsample of coded texts. We then aggregate crowd codings at the text-level using a Bayesian measurement model that leverages the information provided by expert reviews to ensure that the posterior classification of texts aligns with our post-hoc gold-standard judgments. This also drives the measurement model to adjust for judgments of coders’ whose decisions do not align with gold-standard labels, so that the measurement error resulting from low-quality crowd judgments is reduced.
In our working paper Opportunity structures and parties’ anti-elite strategies in multi-party competition, we apply this measurement strategy to explain the anti-elite strategies of parliamentary parties in 14 Western European countries since 2008.