Assessment of Respondent-Driven Sampling Bootstrap Variance Estimators

• Activity Number: 334 - New Methods for Survey Weighting and Variance Estimation
• Type: Contributed
• Date/Time: Wednesday, August 5, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Survey Research Methods Section
• Author(s): Isabelle Beaudry* and Miles Q. Ott
• Companies: Pontificia Universidad Católica de Chile and Smith College
• Keywords: Respondent-Driven sampling; Survey sampling methodology; Hard-to-reach populations; Network inference; Social networks; Bootstrap variance estimators

Abstract

Respondent-Driven sampling (RDS) is a sampling method designed to access hard-to-reach populations by leveraging social connections among the members of the target population. Substantial statistical advancement has been made to estimate the prevalence of an outcome variable with RDS data over the past two decades. In addition, bootstrap-based procedures have been introduced to estimate the variance of these prevalence estimators. However, with a few exceptions, the performance of the bootstrap procedures under various network and sampling conditions has not been the subject of comparative studies. Consequently, the purpose of this work is to present a comprehensive assessment of multiple RDS bootstrap procedures. In particular, we assess the Salganik (2006), the Successive Sampling (2011), the Lu (2013), the Yamanis, et al. (2013), the Model-Assisted (2015), and the Tree Bootstrap (2016) variance estimators via a Monte Carlo simulation designed to capture various levels of network homophily, tie reciprocity, seed selection bias and differential recruitment. Finally, the variance estimators are also compared in real network datasets.