Model-Based Estimation for Respondent-Driven Sampling

• Activity Number: 80
• Type: Contributed
• Date/Time: Sunday, July 29, 2012 : 4:00 PM to 5:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Sergiy Nesterko*+ and Joseph Blitzstein
• Companies: Harvard University and Harvard University
• Keywords: respondent-driven sampling ; networks ; homophily ; estimation

Abstract

Respondent-Driven Sampling (RDS) is a standard process for sampling from hard-to-reach populations such as injection drug users or men who have sex with men, widely employed by public health agencies around the world. RDS recruits respondents by tracing links in the network of connections of individuals from the underlying hidden population, with current participants financially encouraged to refer next ones. Here we develop the first fully model-based procedure for RDS to estimate population averages, as well as its extensions to handle missing data and different participant types. The approach does not model the underlying network of individuals, and introduces dependence in observations by assuming homophily, i.e. the tendency of individuals with similar measurements to connect more often. Via simulation, we show the method's superiority in terms of uncertainty intervals coverage rates and mean squared error (MSE) when comparing to the current estimation techniques. We also demonstrate an application of our method to a survey of individuals at high risk of HIV conducted in San Diego.