Respondent Behavior in Respondent-Driven Sampling: Sensitivity Analysis Based on Real Network Data

• Activity Number: 261
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Graphics
• Author(s): Isabelle Beaudry*+ and Krista J. Gile and Corinne M. Mar
• Companies: and University of Massachusetts, Amherst and University of Washington
• Keywords: Network Inference ; Respondent-driven Sampling ; hard-to-reach populations ; AddHealth ; social networks ; network sampling

Abstract

Respondent-driven sampling (RDS) is designed to study hard-to-reach populations. Beginning with a convenience sample, each participant receives a small number of coupons, which they distribute to their contacts who become eligible. Such a sample has the advantage of limiting the dependence of the final sample on the initial sample. However, this method may also lead to inferential bias and elevated variance in many realistic settings. Extensive research has been done using simulation studies to measure the sensitivity of RDS estimators to various sampling and network features. However, most previous work has used only simulated networks, addressed only isolated features of estimators, or has been done on single data set. The main objective of this study is to compare the performance of different estimators under several sampling conditions and across many real networks with varying features. In particular, we study conditions of respondent referral bias, differential recruitment effectiveness, and initial convenience samples, in the context of many real networks from the National Longitudinal Study of Adolescent Health.