Computational ethnography and public health: Scaling and deepening lived experience research on social determinants of health with large language models

This paper examines whether large language models (LLMs) can address a tension in public health between ethnographic research on lived experiences (obtained, for example, through participant observation) and scalable interventions based on quantitative analysis. It puts forward the idea of "computational ethnography" as a potential way to bridge this divide. Computational ethnography is an emerging research paradigm exploring how computational tools, including natural language processing tools like LLMs, might complement traditional ethnographic workflows to improve insights around socio-cultural phenomena including lived experiences of health. This paper reflects on the theoretical and practical potential of LLMs for such purposes, focusing on two key applications: LLM-assisted interviewing and LLM-based analysis. It suggests that LLMs offer opportunities for both scaling and deepening ethnographic research by supplementing interviewing capacities and providing computational assistance for inductive, deductive, and abductive processes of ethnographic analysis. In doing so, it offers potential to identify social determinants of health that traditional population-level studies miss in structured datasets while preserving the contextual understanding that characterises small-scale qualitative studies. However, realising this possibility requires addressing significant ethical, epistemic, and resource challenges that warrant further investigation by the public health community.