Research
Chris Soria is a PhD candidate in Demography at the University of California, Berkeley, where he studies how social networks shape health and cognitive aging across diverse populations. His NIH/NIA F31-funded dissertation examines how social network characteristics influence cognitive decline in older adults. He brings rigorous computational and AI methods to bear on substantive questions about networks, health, and aging. His work has been published in the American Journal of Epidemiology, BMC Geriatrics, and The International Journal of Aging and Human Development, and presented at the Population Association of America (PAA), the American Public Health Association (APHA), and the Pacific Sociological Association (PSA).
Networks, Health & Aging
My primary research examines how social network structure and quality influence health and cognitive aging at both individual and population scales.
Social isolation, loneliness, and cognitive decline. Using longitudinal data from the Health and Retirement Study, I investigate how objective social isolation and subjective loneliness independently shape cognitive trajectories in older adults, with particular attention to post-COVID dynamics. This work is funded by an NIH/NIA F31 fellowship.
New tie formation after network shocks. Building on the concept of social network cognitive buffers, I examine whether restoring lost connections or forming new ties after network disruptions can improve wellbeing and self-rated health.
Social networks and mortality. In collaboration with Dennis Feehan, I use measures from 21 billion Facebook friendships to examine how county-level network cohesiveness and diversity relate to U.S. mortality disparities. Network structure rivals traditional predictors like smoking and income in its association with mortality. Preprint
Dementia classification across diverse settings. I contribute to work assessing the 10/66 algorithm for dementia classification, mapping methods validated in low- and middle-income countries to U.S. data—bridging my substantive interest in dementia with my methodological focus on algorithmic classification. Published in the American Journal of Epidemiology.
Caribbean American Dementia and Aging Study (CADAS). I am part of the CADAS research team, a population-based study of aging and dementia in Cuba, the Dominican Republic, and Puerto Rico. Published in BMC Geriatrics.
Partisanship & Health
Partisan differences in health behaviors and respiratory disease dynamics. With Audrey Dorelien, Ayesha Mahmud, and Dennis Feehan, I examine how partisan affiliation shapes health behaviors and contact patterns, and how these differences alter respiratory disease dynamics. Using data from the Berkeley Interpersonal Contacts Study, we show that Republicans report 20% more daily contacts, adopt protective behaviors at lower rates, and exhibit strong political homophily. Our simulations demonstrate that network structure can more than double the infection gap between partisan groups and shift epidemic peaks by up to 46 days. Preprint
Computational & AI Methods
I develop computational tools and methods that serve my substantive research on networks, health, and aging.
CatLLM. CatLLM is an open-source Python package and web app I created for applying language and vision models to survey coding, image analysis, and data categorization. It emerged from challenges I encountered analyzing open-ended survey responses in my own research. PyPI · GitHub · Web App
High Agreement, Different Stories. My research demonstrates that leading LLMs achieve 88–97% agreement with expert human coders across survey categorization tasks, while revealing systematic differences in how models and humans handle ambiguity. Under review at JSSAM.
Transforming dementia research through algorithmic classification. Extending my computational methods to dementia research, I develop approaches that use algorithmic classification to improve the consistency and scalability of dementia assessment across clinical settings—bridging my methodological and substantive research streams.