Bivariate Association in Respondent-Driven Sampling Data

• Activity Number: 405 - Nonparametric Testing in Complex Data Settings
• Type: Contributed
• Date/Time: Tuesday, August 1, 2017 : 2:00 PM to 3:50 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Dongah Kim* and Krista Gile and Honoria Guarino and Pedro Mateu Gelabert
• Companies: University of Massachusetts, Amherst and University of Massachusetts and National Development and Research Institutes, Inc. and National Development and Research Institutes, Inc.
• Keywords: Inference ; Dependent variables ; non-parametric testing ; Bivariate associations ; network data ; Respondent Driven Sampling

Abstract

Respondent-Driven Sampling (Heckathorn 1997) is a sampling method designed to collect data from hard-to-reach populations; injecting drug users, and sex workers. Beginning with a convenience sample, the sample recruits other participants using a small number of uniquely-identified coupons to distribute among his/her social network. Coupon recipients also receive a small number of coupons to recruit other participants. Using these processes, the survey team can reach a desire sample size of the target population. This method is very effective to collect a data from hard-to-reach populations. However, valid statistical inference for these kinds of data relies on many strong assumptions. Most of all, statistical tests for between pairs of variables has strong limitations. In standard survey samples, we can assume observations from pairs of individuals are independent. In RDS, however, this assumption is not satisfied because of the sampling dependence between individuals. Therefore, we propose a method to non-parametrically estimate the null distributions of standard test statistics in the presence of sampling dependence, allowing for more valid statistical testing.