From Theoretical to Real-World Networks: Using Empirical Samples of Female Sex Workers in China to Evaluate Respondent-Driven Sampling

Giovanna Merli, Duke University
James Moody, Duke University
Jing Li, Duke University
Jake Fisher, Duke University
Sharon Weir, University of North Carolina at Chapel Hill
Xiangsheng Chen, National Center for STD Control, China

Respondent Driven Sampling (RDS) is an increasingly popular chain referral sampling approach with the aim to provide a probability-based inferential structure for representation of hidden populations. Its ability to make inference to the population rests on stringent theoretical assumptions about the referral practices and the structure of the underlying social network that are not observed. Here we take advantage of multiple observation schemas of female sex workers in China to explore referral bias in empirical Respondent Driven samples. We use information not typically collected in standard RDS protocols to (a) assess the validity of RDS participants’ networks self-reports and corroborate reports on network alters’ attributes and relationship attributes; (b) model the recruitment process through dyad-level logistic choice models of recruitment to characterize the mixing patterns of recruitment and identify sources of bias in RDS recruitment; (c) gauge the impact of recruitment bias on the RDS estimates.