Bayesian Resolution of Discrepant Self-Reported Network Ties
Abstract: Most social network analysis assumes an objective network of shared social ties, typically measured as self-reports from research subjects. Although it is common for two parties to give discrepant reports of their shared relationship status, there is no standard way to resolve such discrepancies. We develop a Bayesian model that leverages patterns of agreement among respondents across multiple relations, using flexible priors to allow for aberrant reporting behaviors. The model allows for posterior inference for individual reporter error rates and for the underlying true network. The method is motivated by and applied to the Food, Activity, Screens, and Teens (FAST) study, an investigation of social networks and health behavior among U.S. middle school students.
This CSSI-inspired work is joint with Maryclare Griffin, Dongah Kim, James Kitts, David Nolin, and John Sirard.
Krista Gile earned her PhD in Statistics from the University of Washington in 2008, completing the Social Science Statistics Track. After a postdoc at social science Nuffield College, Oxford, she joined UMass in 2010 as part of the initial cluster in computational social science. Her research focuses on developing statistical methodology for social and behavioral science research, particularly related to making inference about hard-to-reach human populations and from partially-observed social network structures. Much of her work is focused on understanding the strengths and limitations of data sampled with link-tracing designs such as snowball sampling, contact tracing, and respondent-driven sampling.
