Research

Many natural disasters central to climate-policy debates are hydro-meteorological hazards, yet it remains unclear how global media attention is distributed across disaster types. 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 reporting following natural disasters. Cross-border attention rises after disasters but is highly uneven across hazards, with the largest short-run increases following earthquakes (b = 0.0785, 95% CI [0.0758, 0.0812]), dry-mass movements (b = 0.0531, 95% CI [0.0349, 0.0713]), and volcanic eruptions (b = 0.0425, 95% CI [0.0359, 0.0490]). In contrast, climatologically salient hazards such as floods (b = 0.0069, 95% CI [0.0062, 0.0076]) and droughts (b = 0.0001, 95% CI [−0.0036, 0.0039]) receive substantially less coverage. Conditional on severity and duration, hydro-meteorological (“climate-linked”) disasters receive less cross-border attention than geophysical disasters (θ = −0.0065, 95% CI [−0.0115, −0.0015]). Attention also increases with fatalities: disasters with 100+ deaths receive more coverage than those with 0–9 deaths (b = 0.0360, 95% CI [0.0231, 0.0490]), and this fatality gradient is stronger for country pairs with tighter social ties and deeper ancestral links. These patterns highlight systematic cross-border differences in attention to disaster risks.

Objectives: Large language models (LLMs) are increasingly used for health advice, but evidence on how performance varies across languages, topics and sources remains limited. We assess linguistic and contextual variation in Artificial Intelligence (AI) health-claim verification. Methods: We evaluated seven widely accessible LLMs using UK and European Union (EU) health-claim registers and 9088 journalist-vetted PUBHEALTH claims. The register-language analysis used authorised entries translated into 21 languages; status-labelled UK/EU extensions added non-authorised entries for diagnostic checks. PUBHEALTH claims spanned COVID-19, abortion, politics and general health across government, media and other source domains, including nature.com. Models classified claims as supported or unsupported using modal responses across repeated runs. Results: Authorised-claim sensitivity was highest in English and closely related European languages and declined with syntactic distance from English. In status-labelled UK and EU register extensions, models recognised authorised claims but often overclassified non-authorised claims as supported. PUBHEALTH accuracy was lower and varied by topic and source; inaccuracy was highest for general-health claims and claims linked to nature.com. Discussion: High authorised-claim sensitivity for English register claims can hide lower sensitivity in other languages and weaker performance in other public-health contexts. Training-data exposure, editorial framing and topic-specific tuning are plausible contributors to these gaps. Conclusions: LLM performance in health-claim verification varies substantially by language, topic and source. English-language performance does not reliably generalise across contexts, underscoring the need for multilingual, domain-specific evaluation before public-health deployment.

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.

Social media platforms mediate professional communication, political expression, and community formation, making the rare instances when users collectively abandon an incumbent platform particularly consequential. Strong network effects raise switching costs and strengthen incumbents' positions, making coordinated exit difficult. Here we link 276,431 scholars on Twitter/X to their respective new profiles among the universe of all 16.7 million Bluesky accounts, tracked from January 2023 to December 2024, using a scalable, high-precision cross-platform matching pipeline. Exploiting exogenous variation from Brazil's court-ordered suspension of Twitter/X and a dynamic matching design, we show that adoption is peer-driven, treatment effects are short-lived and dose-dependent, and contagion is simple, not complex. Three patterns characterize adoption and retention. Adoption concentrates among users deeply embedded in Twitter's social graph. Public political expression predicts migration, consistent with homophilous inflows into a largely left-of-center Bluesky information space. Early reconnection with prior contacts predicts longer tenure and engagement. Our findings provide the first population-scale causal evidence of peer influence in a social media platform migration by exploiting exogenous exposure variation in a natural experiment and using daily dynamic matching. Rather than the complex contagion mechanism often emphasized in the literature, contagion is predominantly simple. Our findings recast migration as a multi-homing strategy that insures against governance uncertainty and show that users who quickly reconnect with prior contacts remain active longer on Bluesky.

We examine whether research systems reallocate scientific effort as health needs change. We assemble a global disease–location panel for 204 countries and territories (1990–2021) by linking disease-specific publication output to disease burden in the same place and year. Using large language models, we extract diseases from article text, map them into a disease classification, and classify research funders by type. We estimate how publication output co-moves with disease burden within countries and diseases over time, and use event-study designs around the timing of outbreak alerts, including models with alerted-cell pre-outbreak trends. We find that responsiveness to endemic burden has increased over time but remains highly uneven across locations; outbreak alerts are followed by rapid, statistically significant research surges concentrated after 2000; and funder portfolios differ sharply in the diseases they support, with philanthropic funders tilted toward several diseases concentrated in lower-income settings and corporate funders tilted toward major non-communicable diseases.

Automation affects the labour content of work differently across different contexts. Yet, most existing exposure measures assign fixed scores to tasks or occupations, limiting comparisons of automation exposure across countries. We develop a task-based and country-specific approach to classify automation exposure across the world to disentangle labor-substituting from labor-augmenting automation, the relevant technology channel, and the material role of AI. Our measure spans 124 countries, generating an atlas of 2.33 million task-country labels for economies covering 99% of world population and GDP. We present five descriptive results. First, exposure is highly uneven, ranging from 3.3% of tasks in South Sudan to 61.6% in China, and rises strongly with income, although substantial variation remains within income groups. Second, across countries, exposed tasks are skewed towards substitution rather than augmentation, but low-income countries are disproportionately exposed to substitution, whereas middle-income countries are more heterogeneous. Third, less technologically advanced forms of automation account for more than half of exposed tasks in low-income countries but about one quarter in high-income countries; while other more complex channels generally rise with income levels. Fourth, AI tends to be less prevalent in simpler channels of automation, but also more prevalent in labour-substituting margins in lower income settings and to augment labour in higher income settings. Fifth, we find that females seem to be disproportionately more exposed to labour-substituting automation than males. Our methodology provides a basis for comparing automation exposure across development stages, linking it with cross-country data and allowing us to treat exposure levels, labour margins, technological channels and AI involvement as separate dimensions.

As economics scales, a key bottleneck is representing what papers claim in a comparable, aggregable form. We introduce evidence-annotated claim graphs that map each paper into a directed network of standardized economic concepts (nodes) and stated relationships (edges), with each edge labeled by evidentiary basis, including whether it is supported by causal inference designs or by non-causal evidence. Using a structured multi-stage AI workflow, we construct claim graphs for 44,852 economics papers from 1980–2023. The share of causal edges rises from 7.7% in 1990 to 31.7% in 2020. Measures of causal narrative structure and causal novelty are positively associated with top-five publication and long-run citations, whereas non-causal counterparts are weakly related or negative.

This paper pilots a method for measuring firms’ product and process innovation at scale. Our approach uses LLMs to complete a detailed innovation survey for a sample of 600,000 companies operating across 40 countries. The preliminary results contain 6 million individual instances of innovative activity along with rich textual descriptions. These responses carry signal, for example a one-standard-deviation increase in a firm’s product innovation index is associated with a 5% higher measured TFP residual. The data reveal that innovation is widespread: two-fifths of companies engage in product innovation and one-quarter in process innovation, far exceeding the share of patenting firms. Firms are more likely to have labor-augmenting than labor-saving introductions, while about 14% have green product innovation and 12% have green process innovation. Greenness and net labor augmentation are positively associated across countries. Beyond our specific application, this paper illustrates the potential for using LLMs to generate survey-level insights at population scale.

Choosing what to work on is one of the least formalized decisions in science. I study whether the local structure of past research helps screen which still-open questions are worth reading first. I build a directed literature graph from 242,595 published economics-facing papers from 1976 to 2026, rank open questions using only the literature available at each date, and test those rankings against which questions later enter published work. On realistic shortlists, graph-based screening outperforms ranking by popularity alone, and a second-stage model built on the same graph improves further. The gains differ across two broad kinds of question. Questions that deepen existing claims with clearer mechanisms produce more later realizations per inspected shortlist, while questions that close missing direct relations cover a larger share of the links the literature eventually adds. More broadly, the historical record suggests that economics more often adds mechanisms around existing claims than closes locally implied direct relations. The paper therefore contributes in two ways. Practically, it shows that local literature structure contains useful screening information. Descriptively, it uses that screen to characterize how economics more often moves.

As social media becomes prominent within academia, we examine its reputational costs for academics. Analyzing Twitter posts from 98,000 scientists (2016–22), we uncover substantial political expression. Online experiments with 6,000 U.S. respondents and 135 journalists, rating synthetic academic profiles with different political affiliations, reveal that politically neutral scientists are seen as the most credible. Strikingly, political expressions result in monotonic penalties: Stronger posts reduce perceived credibility of scientists and their research and audience engagement more, particularly among oppositely aligned respondents. Two surveys with scientists highlight their awareness to penalties, their perceived benefits, and a consensus on limiting political expression outside their expertise.

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.

For six decades, Bob Dylan has challenged listeners with songs that reward interpretation. Critics and fans have long pored over his words, treating them as literary texts worthy of a slow, devotional reading, line by line, image by image. In 2016, Dylan even won the Nobel Prize in Literature. As the Swedish Academy put it, the prize honoured him for 'having created new poetic expressions within the great American song tradition'. But what more might we discover if, instead of a human scholar, we asked an artificial intelligence to sift through every word Dylan ever wrote? What patterns, connections or evolution in Dylan's massive body of lyrics might reveal themselves to a machine's analysis, and what could that tell us about the man and his music?