![IconGems-Print](images/IconGems-Print.png)
A Comparison of Estimation Methods for Web-Based Respondent Driven Sampling
Nada Ganesh
NORC at the University of Chicago
Stuart Michaels
NORC at University of Chicago
Vicki J. Pineau
NORC at University of Chicago
Becky Reimer
NORC at University of Chicago
Kanru Xia
NORC at University of Chicago
Traditional probability-based sampling strategies are impractically expensive for surveying rare populations. As a result, rare or hidden populations are often studied using versions of convenience samples, for which study results not necessarily projectable to the target population. In an effort to improve methods for surveying rare or hidden population, NORC conducted a pilot study which tested a cost-effective alternative that combines probability sampling and Respondent-Driven Sampling (RDS) which we refer to as Web-based RDS. The initial, or seed, probability sample source for the pilot study is NORC's AmeriSpeak Panel®. The non-probability sample in the pilot study is generated from the seed sample using RDS methods in which respondents refer friends and family to take the survey. We will present results from applying alternative estimation techniques to the Web-based RDS sample obtained in the pilot study, comparing the point estimates and their associated variances for health outcomes under each alternative to each other. We will also compare estimates using an RDS estimator to the alternative estimators that combine probability and non-probability based samples.