Exploring Error Properties of Respondent-Driven Sampling

• Activity Number: 44
• Type: Contributed
• Date/Time: Sunday, July 29, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Sunghee Lee*+ and Tuba Suzer-Gurtekin and James Wagner and Richard Valliant
• Companies: University of Michigan and University of Michigan and University of Michigan and University of Maryland
• Keywords: Respondent driven sampling ; rare population

Abstract

Respondent driven sampling (RDS) is widely practiced for studies of hard-to-reach, hidden or elusive populations who cannot be easily sampled with traditional probability sampling approaches. By utilizing respondents as recruiters of subsequent respondents, RDS claims to produce representative samples after multiple waves. However, this claim requires meeting three critical assumptions based on incentive systems, first-order Markov chains, and network theory for modeling biased selections. Because it is difficult to test whether these assumptions are satisfied and because there is no viable external criterion to incorporate, error properties of RDS estimates have been under-examined. Moreover, most research data collected through RDS are not publicly available. Therefore, the survey methodology field has not seen an active assessment of RDS and its claimed representativeness. This study attempts to examine recruitment processes of RDS as a source of potential error using Sexual Acquisition and Transmission of HIV Cooperative Agreement Program, the only publicly available data source using RDS.