Network Sampling and Subsampling

Joe Blitzstein

Random graphs are increasingly being used as models for social, biological, and information networks, particularly as a null distribution against which to compare hypotheses observed in a real network. As the complexity of such a model is increased, the computational difficulties associated with generating from and fitting the model increase rapidly.

Coming from another direction, there is also great interest in estimating quantities within a network, based on observing a relatively sparse subnetwork. This has led to the study of link-tracing methods such as snowball sampling and respondent-driven sampling. We will describe ways in which these two problems are related, and discuss Monte Carlo and iterative strategies for simulating estimating the parameters of random network models.