Online Program

Using Multiple Data Sources to Explore Referral Bias in Respondent Driven Sampling

*M. Giovanna Merli, Duke University 
James Moody, Duke University 
Whipple Neely, Independent Consultant 
Jing Li, China National Center for STD Control 
Sharon Weir, University of North Carolina 
Xiangsheng Chen, China National Center for STD Control 
Jake Fisher, Duke University 

Keywords: social networks, sampling, RDS, female sex workers, China

Respondent Driven Sampling is an increasingly popular method to recruit samples of hidden populations with the aim to provide a probability-based inferential structure for representations of these populations. The validity of the RDS estimates of characteristics of the hidden population rests on stringent theoretical assumptions about the referral practices of participants to new participants and the structure of the underlying social network. We take advantage of unique information, not typically collected or utilized in standard RDS protocols, on the attributes of respondents’ network alters and of the relationship between respondents and their alters. This information enables (a) an assessment of respondents’ reports on network alters’ attributes and relationship attributes; (b) the modeling of the recruitment process through dyad-level logistic choice models of recruitment to characterize mixing patterns of recruitment and identify sources of recruitment bias; (c) the quantification of the amount of bias in the RDS estimates based on a comparison of networks consistent with expected patterns of recruitment in line with RDS assumptions and with actual recruitment patterns.