Using Multiple Data Sources to Improve Respondent Driven Sampling Estimation

摘要：本研究旨在对中国女性性工作者这一隐蔽人群的性传播疾病（STD）感染、危险和预防性健康行为的患病率进行有效估计。我们利用多种观察模式来提高被调查者驱动抽样（RDS）的实用性。RDS是一种越来越受欢迎的抽样方法，用于招募隐藏人群的样本，目的是为诸如注射吸毒者、性工作者、男男性行为者以及身份特征不太可能被综合调查研究揭示的人口群体的代表性提供基于概率的推断结构，因为他们是罕见的、社会污名化的和/或非法的。RDS利用隐藏人群的社会网络结构来识别和采访参与者。它的有效性取决于严格的理论假设，即参与者向新参与者的推荐实践以及未被观察到的潜在网络结构。尽管CDC和类似的组织对RDS进行了大量投资，但我们对其在保持其代表性承诺方面的有效性进行了很少的实证评估。在此，我们建议通过将对现实世界转诊过程的考虑从理论领域转移到实证领域来改善中国女性性工作者代表的RDS。我们对最近在上海和广西柳州进行的两项RDS研究和基于地点的抽样方法收集的数据进行了分析，从而实现了这一目标。我们使用这些重叠的数据收集来观察嵌入在RDS招聘过程中的社会网络信息，并真实地模拟RDS设置，以便开发改进的RDS估计，以适应观察到的网络推荐过程。我们为使用RDS方法的研究人员提炼了一些指导方针，以改进RDS估计对其他隐藏人群的表示。

DESCRIPTION (provided by applicant): Using Multiple Data Sources to Improve Respondent Driven Sampling Estimation ABSTRACT: This study addresses efforts to obtain valid estimates of the prevalence of sexually transmitted disease (STD) infection and risky and preventive health behaviors in a hidden population, female sex workers in China. We take advantage of multiple observations schemas to improve the utility of Respondent Driven Sampling (RDS). RDS is an increasingly popular sampling method used to recruit samples of hidden populations with the aim to provide a probability-based inferential structure for representations of populations such as injection drug users, sex workers, men who have sex with men and population groups whose status characteristics are not likely to be revealed by omnibus survey research because they are rare and socially stigmatized and/or illegal. RDS capitalizes on the social network structure of the hidden population to identify and interview participants. Its validity rests on stringent theoretical assumptions about the referral practices of participants to new participants and the structure of the underlying network that are not observed. Despite significant investments by CDC and similar organizations in RDS, we have few empirical evaluations of its effectiveness at keeping its representation promise. Here we propose to improve RDS for representation of female sex workers in China by moving considerations regarding real-world referral processes from the theoretical to the empirical realms. We accomplish this with a combination of analyses of data we have recently collected through two RDS studies and a venue-based sampling approach in Shanghai and Liuzhou (Guangxi Province). We use this overlapping data collection to observe the social network information embedded in the RDS recruitment process and to realistically simulate RDS settings in order to develop improved RDS estimates adaptive to the observed network referral process. We distill guidelines for researchers using RDS methods on needed steps to improve RDS estimation for representation of other hidden populations.