Academics have traditionally played a vital role in both the generation and dissemination of knowledge, ideas and narratives. Social media, relative to traditional media, provides for new and more direct ways of science communication. Yet, since not all academics may engage with social media, the sample that does so may have an outsize influence on shaping public perceptions of academia more broadly through at least two channels: the set topics they engage with and through the particular style and tone of communication. This paper describes patterns in academics' expression online found in a newly constructed global dataset covering over 100,000 scholars linking their social media content to academic record. We document large and systematic variation in politically salient academic expression concerning climate action, cultural, and economic concepts. We show that these appear to often diverge from general public opinion in both topic focus and style. 

Support for right-wing populist parties is characterised by considerable regional heterogeneity and especially concentrated in regions that have experienced economic decline. It remains unclear, however, whether the spatial externalities of local decline, including homelessness and crime, boost support for populist parties, even among those not directly affected by such decline. In this paper, we contribute to filling this gap in two ways. First, we gather novel data on a particularly visible form of local decline, high-street vacancies, that comprise 83,000 premises in England and Wales. Second, we investigate the influence of local decline on support for the right-wing populist UK Independence Party (UKIP) between 2009 and 2019. We find a significant positive association between high-street vacancy rates and UKIP support. These results enhance our understanding of how changes in the lived environment shape political preferences and behaviour, particularly in relation to right-wing populism. 

Climate change is increasing the frequency and severity of natural disasters worldwide. Media coverage of these events may be vital to generate empathy and mobilize global populations to address the common threat posed by climate change. Using a dataset of 466 news sources from 123 countries, covering 135 million news articles since 2016, we apply an event study framework to measure cross-border media activity following natural disasters. Our results shows that while media attention rises after disasters, it is heavily skewed towards certain events, notably earthquakes, accidents, and wildfires. In contrast, climatologically salient events such as floods, droughts, or extreme temperatures receive less coverage. This cross-border disaster reporting is strongly related to the number of deaths associated with the event, especially when the affected populations share strong social ties or genetic similarities with those in the reporting country. Achieving more balanced media coverage across different types of natural disasters may be essential to counteract skewed perceptions. Further, fostering closer social connections between countries may enhance empathy and mobilize the resources necessary to confront the global threat of climate change.

Using basic health statements authorized by UK and EU registers and ~9,100 journalist-vetted public-health assertions on topics such as abortion, COVID-19 and politics from sources ranging from peer-reviewed journals and government advisories to social media and news across the political spectrum, we benchmark six leading large language models from in 21 languages, finding that—despite high accuracy on English-centric textbook claims—performance falls in multiple non-European languages and fluctuates by topic and source, highlighting the urgency of comprehensive multilingual, domain-aware validation before deploying AI in global health communication.

This paper leverages generative AI to build a network structure over 5,000 product nodes, where directed edges represent input-output relationships in production. We layout a two-step ‘build-prune’ approach using an ensemble of prompt-tuned generative AI classifications. The ’build’ step provides an initial distribution of edge predictions, the ‘prune’ step then re-evaluates all edges. With our AI-generated Production Network (AIPNET) in toe, we document a host of shifts in the network position of products and countries during the 21st century. Finally, we study production network spillovers using the natural experiment presented by the 2017 blockade of Qatar. We find strong evidence of such spill-overs, suggestive of on-shoring of critical production. This descriptive and causal evidence demonstrates some of the many research possibilities opened up by our granular measurement of product linkages, including studies of on-shoring, industrial policy, and other recent shifts in global trade.

The study measures scientists’ polarization on social media and its impact on public perceptions of their credibility. Analyzing 98,000 scientists on Twitter from 2016 to 2022 reveals significant divergence in expressed political opinions. An experiment assesses the impact of online political expression on a representative sample of 1,700 U.S. respondents, who rated vignettes with synthetic academic profiles varying scientists’ political affiliations based on real tweets. Politically neutral scientists are viewed as the most credible. Strikingly, on both the ’left’ and ’right’ sides of politically neutral, there is a monotonic penalty for scientists displaying political affiliations: the stronger their posts, the less credible their profile and research are perceived, and the lower the public’s willingness to read their content. The penalty varies with respondents’ political leanings.

We analyse the migration of 300,000 academic users from Twitter/X to Bluesky between 2023 and early 2025, combining rich bibliometric data, longitudinal social-media activity, and a novel cross-platform identity-matching pipeline. We show that 18% of scholars in our sample transitioned, with transition rates varying sharply by discipline, political expression, and Twitter engagement but not by traditional academic metrics. Using time-varying Cox models and a matched-pairs design, we isolate genuine peer influence from homophily. We uncover a striking asymmetry whereby information sources drive migration far more powerfully than audience, with this influence decaying exponentially within a week. We further develop an ego-level contagion classifier, revealing that simple contagion drives two-thirds of all exits, shock-driven bursts account for 16%, and complex contagion plays a marginal role. Finally, we show that scholars who rebuild a higher fraction of their former Twitter networks on Bluesky remain significantly more active and engaged. Our findings provide new insights onto theories of network externalities, directional influence, and platform migration, highlighting information sources’ central role in overcoming switching costs.

Cass Sunstein's essay 'On Bob Dylan' describes Dylan's 'dishabituating' style -- a constant refusal to conform to expectation and a penchant for reinventing his musical and lyrical identity. In this paper, I extend Sunstein's observations through a large-scale computational analysis of Dylan's lyrics from 1962 to 2012. Using o3-mini-high (a large language model), I extract concept-to-concept relationships from the lyrics and construct directed knowledge graphs that capture Dylan's thematic structure. I then quantify shifts in sentiment, metaphorical expression, thematic diversity, and network complexity over time. The results indicate that Dylan's lyrics increasingly rely on metaphor, display an evolving sentiment profile, and exhibit heightened dishabituation -- measured here as a growing variance in the network centrality of key concepts. I also find that references to movement, protest, and mythic imagery fluctuate in ways that align with well-known phases of Dylan's career, reflecting the dynamic and unpredictable quality of his art. These findings not only deepen our empirical understanding of Sunstein's thesis but also introduce a novel computational method for analyzing an artist's evolution-offering broader applicability to the study of cultural and creative change.

This open‑source notebook collection and slides demonstrate two complementary LLM paradigms, retrieval and generation, for turning raw text into structured, research‑ready data.

1. Retrieval notebooks show how to mine large document corpora to extract causal edges, stance labels, demographic attributes and other key fields (e.g., the pipeline powering www.causal.claims).

2. Generation notebooks start from minimal seed prompts and leverage the model’s prior to build production networks, innovation profiles and context‑aware keyword dictionaries (see aipnet.io and www.academicexpression.online).

Across both strands you will find hands‑on modules for prompt engineering, JSON‑schema enforcement, cost‑efficient batch calling, embedding‑based code mapping (HS6 / JEL) and validation routines such as modal voting and cosine sanity checks. By the end, users can scale or adapt each workflow—whether analysing messy policy PDFs or constructing supply‑chain graphs—while keeping costs predictable and outputs auditable.

<Preliminary and Incomplete>

Global supply chains now provide most U.S. drugs, devices, and high-risk foods, yet fewer than 2 percent of foreign plants are inspected annually. I merge FDA audit records from 2014 to 2025 with 53 million import-shipment lines covering 2.4 million global facilities. A staggered difference-in-differences analysis reveals that even a clean audit induces substantial product exits and suppresses new introductions, thereby narrowing product portfolios and shifting output toward closely related product lines. I model a plant’s first regulatory inspection as a belief shock to its perceived regulatory burden, sharply increasing its perceived probability of future inspections. Subsequent formal regulatory violations add little beyond this initial effect. These outcomes illustrate the inherent balancing act policymakers face: tightening enforcement enhances quality but reallocates innovation and increases vulnerability within global supply networks.

<Preliminary and Incomplete>

We characterize the structure of conspiracy theories and investigate the determinants of their popularity by analyzing the universe of posts from a popular social media conspiracy forum between 2008 to 2025. Leveraging advanced large language models, we construct structured representations that systematically capture key events, actors, and their interrelations within each conspiracy theory. These representations enable us to characterize the nature of conspiracy theories at scale. Linking conspiracy characteristics to detailed popularity metrics we establish three main findings. First, richer conspiracies with more connections and nodes, but not those with many interconnections are associated with more engagement, consistent with a preference for conspiracies with good model fit that are harder to falsify. Second, more novel conspiracies are less popular, consistent with an important role of familiarity. Third, the popularity of conspiracies increases with the fraction of truthful claims, consistent with partial truths enhancing perceived plausibility..

<Preliminary and Incomplete>

Medical evidence is overwhelmingly produced in high‑income settings, raising concerns that global research priorities may not align with global health needs. We built a geography‑aware knowledge graph that links clinical articles from top 500 medical journals to the diseases they study, the countries whose data they analyze and the institutional homes of their authors. Large‑language‑model extraction and funder classification allow direct comparison between publication output, national disease burden and source of support across 204 countries and 22 major disease groups. The elasticity of research output with respect to domestic Disability‑Adjusted Life Years has more than doubled since 1990, signalling a steady shift toward needs‑driven science. Yet lower‑income countries still provide <1% of authorship and maternal–neonatal disorders, nutritional deficiencies and several infectious diseases remain under‑studied. Philanthropic funders preferentially target neglected burdens, whereas corporate R&D focuses on profitable chronic conditions. Quasi‑experimental evidence from over 3,000 WHO disease outbreak alerts confirms that sudden health shocks trigger rapid, durable increases in both domestic and global research. Together, these results reveal a scientific system that is becoming more inclusive but remains uneven—and offer a scalable framework for tracking progress in real time.

<Preliminary and Incomplete>

Patent records, long used as the primary measure of corporate innovation, systematically overlook vast swathes of inventive activity, particularly in lower-income economies and green technologies. We quantify this blind spot by extracting innovation signals for 4 million firms using a large-language model (LLM) trained only on firm names and locations. After validating these signals against established firm-level data, we aggregate them to a country–industry grid and compare the incidence of LLM-identified innovators with that of patent filers. The resulting innovation mark-up shows that in lower- and middle-income countries, the median sector hosts roughly eight times more innovative firms than patent data indicate—and even more so for green technologies. This mark-up falls sharply with rising per-capita income and human-capital depth, pinpointing middle-income economies as the epicenter of the patent shortfall. Our results recalibrate cross-country innovation comparisons and demonstrate how LLM-based text mining can serve as a low-cost, scalable complement to official IP statistics.