Allegra H. Fullerton

The integration of large language models (LLMs) into public policy research presents both exciting opportunities and methodological challenges. This research note explores how OpenAI's GPT can be used to semi‐automate the annotation of legislative testimony within the Advocacy Coalition Framework, focusing on emotion‐belief dyads. Building on Emotion‐Belief Analysis, we demonstrate how GPT can assist in identifying these complex constructs under human supervision. Our contributions are threefold: (1) we provide practical guidance for applying LLMs to publicly available textual data, (2) we propose a semiautomated workflow that strengthens conceptual clarity, transparency, consistency, replicability, and accessibility, and (3) we reflect on the ethical and methodological implications of LLM‐assisted research. As LLMs continue to advance, this research note aims to help scholars balance innovation with rigor and integrate these tools responsibly into policy research, offering lessons that extend to the study of frames, discourses, narratives, and other ideational dimensions of policymaking.

Advocacy Coalition Framework (ACF) research has been continuously evolving to improve the understanding of coalition studies. This study aims to critically examine the most common methods for coalition identification in ACF research and to identify strategies to strengthen their clarity and interpretation. We identify and develop ideas around four key steps in using social network analysis (SNA) to study coalitions: collecting and understanding data, choosing a community detection algorithm, applying data transformations, and, most importantly, interpreting community structures. We argue that neither this paper nor any others can provide a definitive approach for understanding advocacy coalitions. Instead, our charge to those using the ACF is threefold. First, recognize the elusiveness of advocacy coalitions and the inherent limitations of any representation. Second, acknowledge that network analysis of advocacy coalitions involves numerous combinatorial choices determined by data characteristics, algorithm selection, and data transformation choices. Third, encourage researchers to understand their data, select among these combinations, interpret results, and effectively communicate the sensitivity and robustness of their analyses.

This study examines the evolution of climate change discourse in the United States Congress from 1987 to 2017, employing natural language processing (NLP) techniques to analyze floor speeches. Using a la carte (ALC) word embeddings, we investigate how Democratic and Republican members of Congress frame climate change, focusing on denialist, fatalist, and solution‐oriented language. Our analysis reveals significant partisan divergences in climate change framing, with Republicans increasingly adopting denialist language, particularly around the term “global warming,” while Democrats maintain a more consistent, solution‐oriented approach. Both parties show a gradual increase in the use of fatalistic framing over time. These linguistic patterns reflect broader political strategies and evolving public discourse on climate change. By quantifying these semantic shifts, the findings contribute to the literature on agenda‐setting and policy framing offering a novel perspective on how political frames are constructed and maintained at the institutional level. This research not only enhances our understanding of climate change politics but also provides a methodological framework for analyzing long‐term trends in political discourse across various policy domains.

Most studies of policy change within the ACF have been descriptive, single-instance explorations that confirm one or more of its four pathways: external shocks, internal shocks, negotiated agreements, and policy-oriented learning. To advance ACF studies of policy change, this chapter takes a different approach to assessing the drivers of change. Rather than looking to confirm one of the four pathways for a single instance of change, we inductively explore what policy actors cite as “antecedents”—or potential influencing, motivating, or driving factors—for multiple policy changes in a textbook subsystem. We identify antecedents through policy actor references in the public discourse related to 77 new or modified policies in the Colorado oil and gas subsystem between 2007 and 2021, drawing on public and government documents, public testimony, and news articles. We then compare the inductively identified antecedents of each policy change with the four pathways of change outlined within the ACF. We find references to internal antecedents were the most commonly mentioned, followed by references to external antecedents. We conclude by discussing the significance of these results for studying policy change, the importance of including both inductive and deductive methods to expand the existing pathways, and guidelines for conducting a medium-sized analysis of policy change over time.

Devil and Angel Shifts are commonly applied concepts in policy research. Using the two concepts in related but different theoretical frameworks – the Advocacy Coalition Framework (ACF) and the Narrative Policy Framework (NPF) – provokes potential challenges of cross‐theoretical dialogue, the establishment of methodological best practices, and mutual research advancement. How do these frameworks conceptualize Devil and Angel Shifts theoretically and how do they empirically study this concept? To what extent do similarities and differences exist that influence comparability and research progress? This paper starts from the foundational roots of the two concepts that were developed to advance the understanding of collective action and inter‐coalition conflict. It then provides a systematic literature review of existing applications that sheds light on conceptualization, measurement, hypotheses, and findings. Drawing from these insights and complementing them with psychological research, it makes an integrated argument that ACF and NPF are developing the same sub‐theory of the Devil and Angel Shifts and offers recommendations for improvements.

Two perceptions can be found in any public debate: first, perceptions of the self and how a person describes their own beliefs and emotions; and second, perceptions of the other and how a person ascribes beliefs and emotions to their allies and opponents. These perceptions of allies and opponents have been topics of study in research on the Devil and Angel Shift, as found in the Advocacy Coalition Framework. The goal of this article is to build on and contribute theoretically and empirically to the study of the Devil and Angel Shift using textual data. This article compares coalition dynamics through the use of “other” emotion and belief statements compared with the traditional “self” narrated emotion and belief statements. This article analyzes the positive and negative emotions and associated beliefs of policy actors engaged in a debate on banning gender affirming care in Arkansas, USA. It finds that advocacy coalitions tend to ascribe more negative emotions or devilize their opponents but tend to use negative and positive emotions at about the same frequency toward their allies. This article offers a broader conceptualization of how to study the Devil and Angel shift, introduces new strategies of Unifying and Acknowledging both allies and opponents, and presents a new avenue of research for scholars interested in applications focusing on coalition dynamics.

As one of the most established theoretical approaches to public policy, the Advocacy Coalition Framework (ACF) has moored most of its theoretical arguments around a textbook policy conflict consisting of two or more advocacy coalitions in a mature adversarial policy subsystem within an advanced polyarchy. This article steps beyond the textbook by introducing deep core coalitions marked by compounding intersectional identities operating at the macro-system. It offers two illustrations of deep core coalitions, one bound by their collective transgender identity and the other by their collective traditionalist identity. Finally, this article concludes with a discussion of what it means for a research program to embrace a diverse research agenda, such as through better linkages with other theoretical approaches, launching more comparative research designs, or, as done here, focusing on a new type of advocacy coalition operating at the macro-system.

Many theories and approaches to policy studies have recently begun to question and research how emotions interact with peoples' understanding and behaviors, especially in policy and politics. This paper builds on and contributes to studying emotions in policy and politics via the advocacy coalition framework (ACF). In applying Emotional‐Belief Analysis, this paper examines the legislative testimony on one of the US' first gender‐affirming care (GAC) bans. It shows that those testifying can be organized in competing advocacy coalitions with distinct emotion‐belief expressions in combination with deep core and policy core beliefs. Moreover, expressions of negative emotions and policy core beliefs display significant and the largest effects in explaining coalition affiliation and shared views of the bill banning GAC. The conclusion summarizes the paper's empirical themes with suggestions for incorporating emotions more into the ACF and the broader policy studies field.