Refusal behavior by Large Language Models (LLMs) is increasingly visible in content moderation, yet little is known about how refusals vary by the identity of the user making the request. This study investigates refusal as a sociotechnical outcome through a counterfactual persona design. Focusing on a Vision-Language Model (GPT-4V), we examine how gendered persona in prompts influence refusal in binary gender classification tasks. We vary gender identity across male, female, non-binary, and transgender personas while keeping the classification task and visual input constant. We find that transgender and non-binary personas experience significantly higher refusal rates, even in non-harmful contexts. Our findings also provide methodological implications for equity audits using LLMs. We underscore the importance of modeling identity-driven disparities and caution against uncritical use of artificial intelligence systems for content coding. This study advances algorithmic fairness by reframing refusal as a communicative act that may unevenly regulate epistemic access and participation.

Argument strength and message persuasiveness are key constructs in message effects research. Yet, researchers still lack a systematic and efficient approach to uncover the “recipe” for these message-level latent features. We applied an agnostic causal machine learning approach that integrates the supervised Indian Buffet Process (sIBP) algorithm with AI-facilitated researcher refinement, train/test set splitting, and crowdsourcing in two multiple-message experiments, each with a large stimulus pool. We conducted message-level analyses on (a) textual tobacco control messages (K = 377) among Chinese men who smoke (N = 1,206) and (b) COVID-19 vaccine promotional messages (K = 1,759) among a national sample of U.S. adults (N = 819). This agnostic approach discovered that messages emphasizing negative health consequences increased argument strength and message persuasiveness. In contrast, politicizing cues reduced the message persuasiveness of social media messages promoting COVID-19 vaccines. We discussed the strengths and limitations of this approach for future message effects research.

Elon Musk’s acquisition of Twitter (now ‘X’) raised concerns about its governance and functioning. Using this as a case study, we propose a framework to analyze the interplay of social identities, perceptions, attitudes, and behavioral responses to major platform policy and ownership changes. Analyzing data from a two-wave nationally representative U.S. panel survey, we found that partisanship, after controlling for gender and race, was consistently associated with various perceptions about Twitter. Democrats evaluated Musk’s takeover and Donald Trump’s account reinstatement more negatively than Republicans. Moreover, we observed ‘lagging resistance,’ or a ‘wait-and-see mindset,’ among users: perceptions of negative impact, disagreement with Trump’s reinstatement, and distrust in Musk were associated with only intentions to reduce long-term Twitter use, not actual use after the takeover. Furthermore, our analysis revealed no clear differences across identity groups in platform migration, despite the tremendous attention it received after Musk’s takeover.

While researchers often study message features like moral content in text, such as party manifestos and social media, their quantification remains a challenge. Conventional human coding struggles with scalability and intercoder reliability. While dictionary-based methods are cost-effective and computationally efficient, they often lack contextual sensitivity and are limited by the vocabularies developed for the original applications. In this paper, we present an approach to construct vec-tionary measurement tools that boost validated dictionaries with word embeddings through nonlinear optimization. By harnessing semantic relationships encoded by embeddings, vec-tionaries improve the measurement of message features from text, especially those in short format, by expanding the applicability of original vocabularies to other contexts. Importantly, a vec-tionary can produce additional metrics to capture the valence and ambivalence of a message feature beyond its strength in texts. Using moral content in tweets as a case study, we illustrate the steps to construct the moral foundations vec-tionary, showcasing its ability to process texts missed by conventional dictionaries and word embedding methods and to produce measurements better aligned with crowdsourced human assessments. Furthermore, additional metrics from the vec-tionary unveiled unique insights that facilitated predicting outcomes such as message retransmission.

This study examines how gender stereotypes are reflected in discourses around controversial science issues across two platforms, YouTube and TikTok. Utilizing the Social Identity Model of Deindividuation Effects, we developed hypotheses and research questions about how content creators might use gender-related stereotypes to engage audiences. Our analyses of climate change and vaccination videos, considering various modalities such as captions and thumbnails, revealed that themes related to children and health often appeared in videos mentioning women, while science misinformation was more common in videos mentioning men. We observed cross-platform differences in portraying gender stereotypes. YouTube’s video descriptions often highlighted women-associated moral language, whereas TikTok emphasized men-associated moral language. YouTube’s thumbnails frequently featured climate activists or women with nature, while TikTok’s thumbnails showed women in Vlog-style selfies and with feminine gestures. These findings advance understanding about gender and science through a cross-platform, multi-modal approach and offer potential intervention strategies.

Twitter enables an online public sphere for social movement actors, news organizations, and others to frame climate change and the climate movement. In this paper, we analyze five million English tweets posted from 2018 to 2021 demonstrating how peaks in Twitter activity relate to key events and how the framing of the climate strike discourse has evolved over the past three years. We also collected over 30,000 news articles from major news sources in English-speaking countries (Australia, Canada, United States, United Kingdom) to demonstrate how climate movement actors and media differ in their framing of this issue, attention to policy solutions, attribution of blame, and efforts to mobilize citizens to act on this issue. News outlets tend to report on global politicians’ (in)action toward climate policy, the consequences of climate change, and industry’s response to the climate crisis. Differently, climate movement actors on Twitter advocate for political actions and policy changes as well as addressing the social justice issues surrounding climate change. We also revealed that conversations around the climate movement on Twitter are highly politicized, with a substantial number of tweets targeting politicians, partisans, and country actors. These findings contribute to our understanding of how people use social media to frame political issues and collective action, in comparison to the traditional mainstream news outlets.

Social bots, or algorithmic agents that amplify certain viewpoints and interact with selected actors on social media, may influence online discussion, news attention, or even public opinion through coordinated action. Previous research has documented the presence of bot activities and developed detection algorithms. Yet, how social bots influence attention dynamics of the hybrid media system remains understudied. Leveraging a large collection of both tweets (N = 1,657,551) and news stories (N = 50,356) about the early COVID-19 pandemic, we employed bot detection techniques, structural topic modeling, and time series analysis to characterize the temporal associations between the topics Twitter bots tend to amplify and subsequent news coverage across the partisan spectrum. We found that bots represented 8.98% of total accounts, selectively promoted certain topics and predicted coverage aligned with partisan narratives. Our macro-level longitudinal description highlights the role of bots as algorithmic communicators and invites future research to explain micro-level causal mechanisms.

Scholars increasingly use Twitter data to study the life sciences and politics. However, Twitter data collection tools often pose challenges for scholars who are unfamiliar with their operation. Equally important, although many tools indicate that they offer representative samples of the full Twitter archive, little is known about whether the samples are indeed representative of the targeted population of tweets. This article evaluates such tools in terms of costs, training, and data quality as a means to introduce Twitter data as a research tool. Further, using an analysis of COVID-19 and moral foundations theory as an example, we compared the distributions of moral discussions from two commonly used tools for accessing Twitter data (Twitter’s standard APIs and third-party access) to the ground truth, the Twitter full archive. Our results highlight the importance of assessing the comparability of data sources to improve confidence in findings based on Twitter data. We also review the major new features of Twitter’s API version 2.