Job Market Paper

Black by Popular Demand: Media, Competition and the Evolution of a Social Norm Latest Draft

Leveraging an exogenous shift in language preferences induced by mid-1960s calls from Black Civil Rights leaders to use the term “Black” instead of “Negro,” this paper examines how economic incentives shape local newspapers’ responses to evolving cultural norms. I introduce two novel datasets—a corpus of 58 million newspaper articles and a digitized database of county-level newspaper circulation from 1964—which I match to create a panel of 450 Southern newspapers spanning the period 1960-1973. I provide evidence that newspapers' demand-side exposure to out-group preferences—proxied by local levels of White racial conservatism—predicts both lower and slower adjustment to the new label, and that this effect is exacerbated by the intensity of local newspaper market competition. I show that the effect is not driven by local journalists, editors or companies' policies. I then, in a second stage, show that, when the ``Black'' label is adopted, media representation of Black entities is significantly, immediately and persistently aligned with the identity intended by the ``Black'' campaign leaders. This representation is less prominent in the face of increasing White racial conservatism and heterogeneous in the face of competition. This paper highlights how economic incentives within media markets can slow the diffusion of evolving norms, particularly when these norms challenge entrenched attitudes within the majority. My findings challenge the conventional view that competition inherently fosters diversity in media content, suggesting instead that it may deepen entrenched biases when minority preferences diverge from majority tastes.

Work in Progress

The Introduction and Reduction of Estimation Bias in AI Assisted Research

When do societal and historical biases introduce measurement error in LLM assisted research? How can we reduce or eliminate these biases? In this paper, I demonstrate how careful methodological design can leverage the power of LLMs while preserving objectivity, thereby enhancing the reliability of AI-assisted textual analysis in economics and the social sciences.

Editorial Bias: Evidence from Historical Newspapers

This paper provides evidence of racial biases in newspaper editor content decisions. It examines subtle variation in language choice across newspapers when publishing the same underlying wire articles, revealing systematic differences in editorial choices.

Selective Science: How Media Bias Shapes Climate Change Narratives (Joint with M. Pograxha)

We investigate how the alignment between scientific narratives in academic articles about climate change and the priors of media consumers influences newspaper coverage. Using a comprehensive dataset of academic papers, media outlets’ political alignments, and metrics of article quality, we explore whether media outlets selectively highlight scientific evidence that confirms their audience’s priors, independent of quality. This research informs our understanding of how media shapes public perceptions of climate science and highlights pathways for improving the dissemination of accurate scientific information.

Moderation of toxic content by downstream supply chains

I explore whether downstream supply chains can impact the moderation of social media platforms and offline behavior, using the case study of multiple alt-right social media platforms.

Development

When to baseline - The impact of lottery allocation household outcomes in a highly vulnerable setting (Joint with C. M. Fernandez, A. Guariso, M. Holmlund, T. Mitchell, and C. Newman.)

It is considered 'good practice' to collect a baseline survey for the purposes of improving statistical power, assessing baseline randomised group balance, and correcting for possible attrition, as well as for the purposes of heterogeneity analysis at end-line. However, it is common in field experiments to randomise treatment status in the public domain to allow community and government scrutiny of the 'fairness' of allocation. This approach can create constraints on the timing of baseline survey implementation. We analyse whether public lottery allocation effects respondents' outcomes in a sample of villages in Niger.