Correcting for differential recruitment in respondent-driven sampling data using ego-network information (Q2192315)

Respondent-Driven Sampling (RDS) is a sampling technique employed to collect data from hard-to-reach human populations. Individuals initially selected are asked to recruit from their contacts in the target population till a pre-assigned sample size is reached. It is normally assumed that recruitment by participants is at random. In practice, this is an unrealistic assumption and the aim of the paper is to introduce three types of non random recruitment and present a framework to estimate sampling probabilities under Differential Recruitment (DR) and provide bias correction for prevalence estimators. Variance estimators are also derived by boot strap methods. DR could be through recruitment based on the recruiter's characteristic (called `nodal attributes') which are categorized as `within group DR' and `between group DR' and a third type based on nature of relationship of recruiter( called `tie attributes'). A simulation study is presented and the proposed methods are applied to data collected by the US Centers for Disease Control and Prevention(CDC) which uses RDS for behavior prevalence surveillance among people who inject drugs (PWID). An appendix at the end gives the stationary distributions with DR, consistency of prevalence estimators, transition probabilities and other mathematical details.