Innovations in the Recruitment of Respondent Driven Samples for Improved Inference to Hidden Populations
Ashton M. Verdery, University of North Carolina at Chapel Hill
Giovanna Merli, Duke University
James Moody, Duke University
Jeffrey Smith, Duke University
Respondent driven sampling (RDS) is a data collection approach for hidden and rare populations which relies on respondents’ social networks to recruit participants and uses post-sample weighting procedures to obtain population representative estimates. RDS’s ease of recruiting study participants quickly and cost-effectively comes at the cost of multiple assumptions about the sampling process which have been shown to be violated in practice with resulting large biases. Whereas prior work to improve the ability of RDS to produce valid and precise estimates of the characteristics of a hidden population has focused on the development of new statistical estimators, we test innovations to the RDS sample recruitment process itself. Using simulated sampling and empirically informed social networks of a hidden population of female sex workers in China, we examine whether modifications to the sampling procedure robustly improve RDS inference across a range of hidden population contexts.