Aaron Danielson

Network Imputation with Copula-based Methods

Network surveys allow social scientists to observe a fraction of the social structure of interest.  When the observed subnetworks are representative of the more extensive network in which they are embedded, analysis of the subnetwork can serve as a proxy for analysis of the entire population.  But, frequently the social scientist is either (1) interested in phenomena related to the network as a singular entity or (2) the subnetwork does not represent the structure in the complete network.  Motivated by a network survey of a large institution in southern California, this talk introduces a novel method for network imputation when the sampled nodes can decide whether or not to respond to the survey.  Specifically, the probability that nodes respond to the survey depends on the decisions made by those to whom they are connected in the network under study.  Copulas provide a natural way to model the network formation and survey response processes.  Beyond this particular application, the method can be used whenever random variables associated with nodes depend upon one another via the network structure.