Assessing respondent-driven sampling
*Matthew Salganik, Princeton University
Keywords: respondent-driven sampling; snowball sampling; social networks
Respondent-driven sampling (RDS) is a recently introduced, and now widely used, technique for estimating the characteristics of hidden populations such as drug injectors and sex workers. RDS data are collected through a snowball mechanism, in which current sample members recruit future sample members. If certain conditions hold about the data collection process and the social network structure of the hidden population, one can make unbiased estimates from this data. However, the accuracy of these estimates has been called into question because 1) even if all the assumptions are met, RDS estimates may have extremely high variance and 2) the necessary assumptions about the sampling process are probably not met. In this talk, we will evaluate respondent-driven sampling in light of these two problems and conclude with suggestions for future research.