|
Abstract:
|
Respondent-driven Sampling (RDS) is a recruitment procedure for surveys of hidden or hard to reach populations. In RDS, enrolled subjects get extra rewards for successful recruitment of their friends who are also in the target network. The quality of statistical estimates from RDS depends on assumptions about the nature of the recruitment process and the social network underlying the target population. Selection bias may affect estimates from RDS surveys since selection of subjects is not controlled by researchers, but by the subjects themselves. In this paper, we characterize RDS procedure under a survival framework, establish a rigorous definition of uniform recruitment and develop statistical estimators and formal tests for recruitment bias when the network of respondents and their alters is fully observed. We apply these tests to a unique RDS study of people who inject drugs in Hartford, CT, USA and find strong evidence to reject the hypothesis of uniform RDS recruitment in this population.
|
