Title: Estimating random graph models from observed subgraphs
Authors: Jonathan Stewart - Florida State University (United States) [presenting]
Abstract: The statistical analysis of sampled network data has been an important and underdeveloped topic in the field of statistical network analysis. We consider the problem of estimating a random graph model from only an observed subgraph. Current approaches make either restrictive assumptions about the dependence structure of the random graph model or require missing data methods that are not scalable to larger graphs. Recent developments are presented that design scalable estimation methodology for estimating parameter vectors of increasing dimension for population models of random graphs with dependent edges based on an observed subgraph. Our methodology designs observation processes that exploit the dependence structure of models in order to ensure sufficient information is contained within the sampled subgraph to facilitate scalable estimation. We show that common observation processes for sampling networks produce observed subgraphs which contain sufficient information for estimating models with triadic dependence. We conclude by elaborating sufficient conditions under which we can obtain non-asymptotic bounds on the statistical error of our estimators.