Title: Learning about network features from respondent-driven sampling data
Authors: Forrest Crawford - Yale University (United States) [presenting]
Abstract: Respondent-driven sampling (RDS) is a link-tracing procedure for surveying hidden or hard-to-reach populations in which subjects recruit other subjects via their social network. RDS recruitment does not follow a pre-specified sampling design, so making population-level inferences from samples can be difficult. We will review the graphical structure of RDS samples and present three methods for learning about network features from RDS data. First, we will show how to reconstruct the recruitment-induced subgraph of surveyed individuals probabilistically. Then, using these results, we will explain the circumstances under which it is possible to separately estimate network homophily (on measured traits in the sample) and preferential recruitment. Finally, we will discuss methods for estimating the size of the hidden population network using parametric and semi-parametric models. We apply each method to an empirical RDS dataset, including a large RDS study of injection drug users in Hartford, Connecticut, USA, in which the true population network is known. Together, these results provide a flexible set of methods that allow researchers to learn about features of a hidden social network via RDS surveys.