Abstract:
|
Collecting data in hard-to-reach populations is often challenging, and researchers require specialized sampling techniques to do so. Respondent-driven sampling (RDS) is one method used in that context. RDS data are sometimes used to estimate the population size. For instance, the recapture phase of a capture-recapture sampling may be collected with RDS. This technique is referred to as the object multiplier (OM) when individuals receive a unique object in the capture phase. The number of objects observed in the RDS sample provides information about the population size if the OM phases are performed independently. However, empirical evidence shows that this assumption may be violated in practice. This work proposes hypothesis testing procedures to detect sample dependence. We consider two approaches: bootstrap procedures and trend tests. We evaluate the methods with simulated networks and show that the trend tests detect sample dependence with high probability. We also discuss an application with OM data. Finally, we present a sensitivity analysis displaying how the sample dependence introduces bias and elevated variance in the Lincoln-Petersen estimator.
|