Working Papers
-
Job market paper
Americans' trust in mass media and other institutions has plummeted in recent years. I argue media and experts appear “out of touch” when economic news becomes disconnected from public experience, a misalignment made particularly salient by online commentary spread on social media and news platforms. During the 2021–2025 “vibecession,” when consumer sentiment fell despite positive economic indicators, I show that online commentary about economic conditions aligned more closely with consumer sentiment than mainstream news. I test the implications using an experiment that varies the tone of both economic news articles and a set of anonymous comments. I find comments substantially shape readers' perceptions of public economic experience. Further, positive articles reduce trust in media and experts among economically dissatisfied readers, while pairing positive articles with negative commentary causes even broader declines in trust. Paradoxically, accurate reporting can reduce institutional trust when citizens perceive elite narratives as disconnected from public experiences.
-
How does news coverage track elite political debate? We examine whether mainstream newspaper coverage of climate change reflects the terms of that debate or filters them. Using a large-language-model pipeline, we extract 176,435 sources from 30,988 climate articles across six major U.S. newspapers (2012–2022) and classify them, alongside 26,868 congressional speeches (1994–2023), along two dimensions: scientific urgency–skepticism and economic cost–benefit. We document a widening politics–coverage gap after 2016. In Congress, scientific and economic arguments remain similarly prevalent, whereas mainstream newspapers increasingly emphasize scientific urgency while sharply reducing attention to the economic consequences of climate policy. By contrast, climate skepticism declines in parallel across both Congress and the press. Linking articles to career histories for 3,688 journalists, we show that this divergence is driven primarily by changes in journalist composition. These findings demonstrate how source selection and newsroom composition shape the political information citizens receive.
-
Keyword dictionaries remain widely used in text analysis for their simplicity, transparency, and replicability. However, sensitivity to keyword selection and inability to account for context introduce measurement error that attenuates estimates and obscures real relationships. We introduce ambiguity-robust dictionaries, a text-as-data method that leverages contextual word embeddings to produce more precise measures while preserving the key advantages of dictionary approaches. Our method generates contextualized embeddings for all keyword instances, then applies constrained fuzzy clustering to determine the degree of each instance's membership in a target cluster defined by a small set of researcher-identified anchor words. This approach allows the construction of ambiguity-robust document level word counts, increasing precision and reducing sensitivity to keyword selection. We demonstrate the method using three applications: UN environmental discourse, ethnic bias in Kenyan judicial politics, and moral language in US Congressional speech. In each application, our method yields substantively more robust and precise estimates than conventional approaches.
-
Social media platforms increasingly use attention signals to rank content. Instead of using network based measures or engagement (ie, likes or retweets), platforms optimize for the amount of time users spend attending to, or consuming content. We propose this leads to a phenomenon called digital rubbernecking, where content that is attention-grabbing because of toxicity or negativity is more likely to be surfaced in platforms that optimize for attention. We test this using a simulated social media environment and custom-built ranking algorithms that either prioritize attention or engagement signals. We hypothesize that algorithms optimized for attention are more likely to surface negative content, which in turn increases users' perceptions of the political extremity of others.
-
We report the first direct comparisons of multiple alternative social media algorithms on multiple platforms on outcomes of societal interest. We used a browser extension to modify which posts were shown to desktop social media users, randomly assigning 9,386 users to a control group or one of five alternative ranking algorithms which simultaneously altered content across three platforms for six months during the US 2024 presidential election. This reduced our preregistered index of affective polarization by an average of 0.03 standard deviations (p < 0.05), including a 1.5 degree decrease in differences between the 100 point inparty and outparty feeling thermometers. We saw reductions in active use time for Facebook (−0.37 min/day) and Reddit (−0.2 min/day), but an increase of 0.32 min/day (p < 0.01) for X/Twitter. We saw an increase in reports of negative social media experiences but found no effects on well-being, news knowledge, outgroup empathy, perceptions of and support for partisan violence. This implies that bridging content can improve some societal outcomes without necessarily conflicting with the engagement-driven business model of social media.
-
Assessing the consequences of online slanted news on political opinions and attitudes requires overcoming well-known challenges in the study of media effects. I present a novel experimental instrument that selectively decreases online exposure to partisan media on Twitter. The instrument comes with a number of advantages over current encouragement designs which tend to find mixed or inconsistent effects. First, I am able to eliminate confounding effects introduced by social media ranking systems by varying partisan media using a web browser extension to manipulate the social media feed on the subject's own device. Second, I examine the effects of decreasing news exposure on pre-existing audiences as opposed to increasing news exposure for non news-consumers. Finally, I also articulate the effect of social context on the persuasiveness of partisan news, an important difference in how partisan news is distributed and consumed on social media platforms.
Peer-Reviewed Publications
-
PS: Political Science & Politics, 2025
This article explores the use of large language models (LLMs), specifically GPT, for enhancing information extraction from unstructured text in political science research. By automating the retrieval of explicit details from sources including historical documents, meeting minutes, news articles, and unstructured search results, GPT significantly reduces the time and resources required for data collection. The study highlights how GPT complements human research assistants, combining automated efficiency with human oversight to improve the reliability and depth of research. This integration not only makes comprehensive data collection more accessible; it also increases the overall research efficiency and scope of research. The article highlights GPT's unique capabilities in information extraction and its potential to advance empirical research in the field. Additionally, we discuss ethical concerns related to student employment, privacy, bias, and environmental impact associated with the use of LLMs.
-
Nature Human Behaviour, 2023
We present the results of a large, US$8.9 million campaign-wide field experiment, conducted among 2 million moderate- and low-information persuadable voters in five battleground states during the 2020 US presidential election. Treatment group participants were exposed to an 8-month-long advertising programme delivered via social media, designed to persuade people to vote against Donald Trump and for Joe Biden. We found no evidence that the programme increased or decreased turnout on average. We found evidence of differential turnout effects by modelled level of Trump support: the campaign increased voting among Biden leaners by 0.4 percentage points and decreased voting among Trump leaners by 0.3 percentage points. An important but exploratory finding is that the strongest differential effects appear in early voting data, which may inform future work on early campaigning in a post-COVID electoral environment. Our results indicate that differential mobilization effects of even large digital advertising campaigns in presidential elections are likely to be modest.