Wang, Peng. Evaluating the reproducibility of respondent-driven sampling: using repeated surveys among people who inject drugs in New Jersey. Retrieved from https://doi.org/doi:10.7282/t3-8r6f-0z37
DescriptionDue to stigma or legal issues, populations with higher HIV risk are often hard to reach, which impedes accurate population estimation of HIV burden. To better sample hard-to-reach populations for HIV surveillance, various sampling methods have been designed and/or used since HIV Epidemic following first reported AIDS cases in 1981. As a sampling method to reduce biases of chain-referral samples and to obtain statistically unbiased estimates of population parameters, respondent-driven sampling has been widely used for hard-to-reach populations in HIV surveillance. However, limited studies have assessed the empirical reproducibility of RDS. This dissertation has three specific aims: (1) to review the development of various sampling methods for hard-to-reach populations in HIV surveillance, and then to critically review the assessment of different sampling methods with a focus on RDS; (2) to assess the reliability of naïve RDS samples; (3) to evaluate the consistency of population parameter estimates based on four RDS estimators (RDS-I, RDS-II, Gile’s SS, and HCG). Five rounds (Range of sample size: 394-524) of the National HIV Behavioral Surveillance – People Who Inject Drugs Surveys in Newark, New Jersey were used. The study shows that relatively good reliability of RDS samples and consistency of RDS-II and Gile’s SS estimators were observed. Future studies should consider the superiority of RDS under a specific situation and conduct diagnostic procedures during the stage of data collection to assure implementation accuracy when RDS is used.