Research about the epistemic crisis has largely treated epistemic threats in isolation, overlooking what they collectively say about the health of news environments. This study integrates the literature on epistemic problems and proposes a broadly encompassing framework that departs from the traditional focus on falsehoods: epistemic vulnerability. This framework is an attempt to more fully capture the erosion of authority and value conferred to political information, which has put stress on the public spheres of many democracies. The study develops the EV index to quantify this phenomenon at the system level in a comparative manner. Using OLS regression, I test the relationships between the EV index and various structural characteristics of political and media systems. Findings are remarkably consistent with established typologies of media systems. Northern European countries exhibit greater epistemic resilience, while the US, Spain, and Eastern Europe are more vulnerable. The study also offers strong evidence that populism, ideological polarization, and political parallelism contribute to higher levels of epistemic vulnerability. Conversely, public media viewership and larger party systems are associated with more epistemically resilient societies.

French news channel CNEWS is regularly compared to Fox News due to presumed ties with the far right, and for promoting conspiracy theories and partisan propaganda. Using data from the ReCitCom project, I propose simple metrics for quantifying the partisanship of news audiences cross-nationally and identifying channels like Fox News in other media systems. I apply them to France and find that CNEWS has the most ideologically radical news audience within the French media ecosystem, and that the pattern is more pronounced with higher frequency of use. Comparatively, I show that the Fox News and CNEWS audiences have an equivalent ideological lean that justifies the comparison between the two outlets at the audience level. The two audiences are also disproportionately receptive to far-right candidates compared to other news audiences and their respective national samples.

We take two approaches to understanding democratically corrosive sentiment (DCS) in the US, which we operationalize in terms of populist attitudes, conspiracy beliefs, and expectation of fraud in the next election. Our first approach is media use, which is not well understood as a correlate of DCS beyond generalities about the harms of social media and partisan news. We distinguish between mainstream news and right-wing media, and between three categories of social media: those facilitating stronger ties among users, those facilitating weaker ties, and extremist Alt-Tech brands. Our second approach to explaining DCS is attitudinal. For this, we introduce a concept called Feelings of Being Devalued (FBD), which we offer as a complement to status threat and sense of material deprivation. Using a survey of our design (N = 2,000) fielded in the US in 2022, we show that: (1) mainstream news use and attention to right-wing media have opposite relationships with DCS; (2) not only Alt-Tech social media but also stronger-tie media such as Facebook are correlated with DCS, while use of weaker-tie social media such as X are uncorrelated in a model with a rich set of controls; and (3) FBD is strongly associated with DCS—more so than right-wing authoritarianism, social dominance orientation, and ideology.

This study explores the synergy between large language models and classical statistics in contemporary data science. In the field of large language models, we find there is no one-size-fits-all model which satisfies the needs of other scientists. There are differences in the soft results which may be a limitation on their application. To analyze these differences and lack of robustness, we propose a robust methodology that integrates classical statistical experimental design principles with these advanced models, aiming to identify statistically significant differences among their outcomes. In particular, an experimental design is presented in which the main factors, levels, treatments and interactions that influence the predictions made by different models of complex natural language processing are identified. The main aim of this research is to better understand the influence of some controlled factors that are used in complex natural language processing models by applying classical statistical techniques, providing a comprehensive perspective on the relative effectiveness of different zero-shot classification models. It aims to offer practitioners insights into when and where certain models may be more or less sensitive, facilitating informed decision-making in applying these advanced language models. Additionally, computational results obtained from a pilot dataset are presented. These results illustrate the entire process of the proposed methodology, highlighting the importance of considering statistical evidence when making decisions.