Modelling Policy Action Using Natural Language Processing: Evidence for a Long-Run Increase in Policy Activism in the UK

Analyzing policymaking using archival evidence is common in qualitative studies, but doing so using text-as-data methods is challenging because commonly used techniques favor the identification of policy areas rather than actions. This article employs natural language processing to evaluate how UK governments describe their actions, using the full sample of official (command) papers they produced between 1974 and 2023. The methodology relies on identifying sentence structure rather than a bag-of-words approach, and on explicitly modelling statements related to policy in terms of subjects, predicates, and objects. The analysis identifies a long-run increase in language referring to active intervention, and a decline in deliberative, analytical, and descriptive language, that began as early as the 1980s. Moreover, the objects of this activist language have shifted from systemic to more personalized over time. The implications of these findings for our understanding of policy developments in the post-1980 era are discussed in the conclusion.