Analyzing Party Competition on Climate Change: A Machine Learning Approach


Climate change is increasingly shaping party competition in Western Europe. Still, previous work has primarily examined environmental protection as a proxy for climate change. Few researchers have made the effort to manually code manifestos to map parties’ attention and positions towards the issue of climate change. Indeed, they have shown that climate change is a distinct issue cutting across traditional party lines and demonstrated the necessity of a reliable measure to analyze contemporary party competition. This article addresses three important shortcomings. First, manual coding of manifestos is hard to replicate and requires many resources for future applications. Second, this approach is not feasible for analyzing political text appearing in higher frequency such as parliamentary speeches, press releases, or social media posts. Third, different conceptualizations complicate comparisons between the existing empirical evidence. We aim to overcome these problems by developing a theoretically sound automated classification process. Following a supervised machine learning approach, we build several binary classifiers allowing us to detect the general topic of climate change, identify political actors’ overall position on climate protection and references to climate change mitigation or adaptation, and differentiate between political demands, blame attribution, and (self-)praise. In the first iteration, we fine-tune transformer models with labeled sentences from parliamentary speeches delivered in the German Bundestag. Our classifiers allow us and other researchers to study German party competition on climate change both in the short- and the long-term. In further iterations, we expand our scope by considering both additional languages and forms of political text.

May 2, 2024 12:00 AM — May 4, 2024 12:00 AM
Vrije Universiteit Amsterdam