This event is cancelled.
The joint Statistics/Economics Seminar speaker for Wednesday, February 13, 2019, is Tyler McCormick, an Associate Professor of Statistics and Sociology at the University of Washington, where he is also a core faculty member in the Center for Statistics and the Social Sciences. He is also a Senior Data Science Fellow and co-lead for Data Science Education & Career Development at the eScience Institute, UW's data science center. Tyler's work develops statistical models that infer dependence structure in scientific settings where data are sparsely observed or subject to error. His recent projects include estimating features of social networks (e.g. the degree of clustering or how central an individual is) using data from standard surveys, inferring a likely cause of death (when deaths happen outside of hospitals) using reports from surviving caretakers, and quantifying & communicating uncertainty in predictive models for global health policymakers. He holds a PhD in Statistics (with distinction) from Columbia University and is the recipient of an NIH Career Development (K01) Award, Army Research Office Young Investigator Program Award, and a Google Faculty Research Award. Tyler currently serves as Editor for the Journal of Computational and Graphical Statistics (JCGS). A complete list of papers is available on his website: thmccormick.github.io.
Talk: Estimating spillovers using imprecisely measured networks
Abstract: Estimating indirect effects (spillovers) requires accurate knowledge of the network of possible influence pathways. In practice, however, networks are often measured with error. Errors arise from sources such as respondents' memory lapses, researchers collecting networks at baseline when connections form during the experimental period, or researchers limiting the number of connections a respondent can nominate. Further, network surveys are often prohibitively expensive, so researchers proxy the network via other methods (e.g. defining a possible spillover pathway if individuals are in the same geographic area, in the same classroom in a school, etc.). In this paper we consider the setting where measured connections are a noisy representation of the true pathways of treatment interference. We show that existing methods yield biased estimators in the presence of this mismeasurement, then develop a class of mixture models that account for missing connections and discuss estimation via the Expectation-Maximization (EM) algorithm. We evaluate its performance by simulating experiments on realistic networks and implement our method using data from two published studies: one where networks are constructed using same class as a proxy, and the other where the number of contacts is censored. In both cases incorporating mismeasurement leads to larger treatment effect estimates. This is joint work with Wesley Lee (UW) and Morgan Hardy (NYU Abu Dhabi).