Evaluating Variance Estimators for Respondent-Driven Sampling

• Activity Number: 361
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Survey Research Methods Section
• Author(s): Michael Spiller and Krista J. Gile* and Mark S. Handcock and Corinne M. Mar and Cyprian Wejnert
• Companies: CDC and University of Massachusetts Amherst and UCLA and University of Washington and CDC
• Keywords: Respondent-driven sampling ; Survey sampling ; Hard-to-reach populations ; Social networks ; Network inference ; Public health

Abstract

Respondent-Driven Sampling (RDS) is a method for sampling hard-to-reach populations. It samples population members via their social networks by having sample members recruit their contacts into the sample. Estimation of population characteristics with RDS data is challenging due to the unobserved population network, and multiple estimators are currently used. Past work has focused on point estimation, and no evaluation of currently used variance estimators exists. We evaluate the performance of RDS variance estimators via simulations of RDS on synthetic networked populations. The networks and RDS sampling processes are based on 40 surveys of injection drug users from CDC's National HIV Behavioral Surveillance system. In these simulations, average design effects (DEs) are lower and average 95% confidence interval (CI) coverage rates are higher than suggested in previous work, with average CI coverage of 93%. However, DE and coverage vary across the 40 sets of simulations, suggesting that the characteristics of a given study should be evaluated to assess performance. We also find that simulation results are sensitive to parameters such as sampling with replacement.