Valid Inference for Respondent Driven Sampling of Hidden Networked Populations

描述（由申请人提供）：被调查者驱动抽样（RDS）可以说是最好和最常用的方法，用于调查难以使用标准概率方法抽样的“隐藏”人群。例子包括注射吸毒者（IDU）、男男性行为者（MSM）和女性性工作者。据估计，2006年美国53,600例艾滋病毒新发感染中，仅两个隐藏人群（男男性行为者和注射吸毒者）的感染就占了69%。我们建议开发统计方法来改进RDS样本的估计，重点是开发工具和方法，这些工具和方法将有助于该领域的研究人员，并为他们提供便利。从统计学的角度来看，RDS是一种旨在获得概率样本的自适应抽样制度。它通常在获取样本时是有效的，但在多大程度上它可以被认为是一个概率样本，具有已知的包含概率，是不清楚的。目前对这些包含概率的估计是有问题的。我们将采用最先进的方法，为RDS开发开源、用户友好的统计软件，举办研讨会，在应用和方法研究人员之间建立合作，传播这些新方法，并提供使用所开发工具的培训。这个项目意义重大，因为它将在隐性人口测量方法和有关科学问题方面取得重大进展。作为第一批与实地研究人员密切合作的统计学家，研究人员具有独特的资格，可以将RDS的统计理解系统化。该项目的创新之处在于，它挑战了RDS推理的现有范式，并提出了一种基于前沿统计思想和模型的新方法。缺乏统计方法来证明RDS的合理性。这个项目将产生一个系统的统计框架，在这个框架内，人们可以了解RDS的优缺点。在建议的工作中，我们将表示RDS程序的复杂性，但也允许对结果推断的不确定性进行量化。RDS统计方法的发展对社会科学和行为科学至关重要。我们将通过面向现场研究人员的开源用户友好软件，传播评估、诊断和量化不确定性的方法。这些方法既适用于未来的区域调查数据，也适用于区域调查调查的庞大现有数据库。

DESCRIPTION (provided by applicant): Respondent Driven Sampling (RDS) is arguably the best and most common method being used to survey "hidden" populations that are hard to sample using standard probability methods. Examples include injection drug users (IDU), men who have sex with men (MSM), and female sex workers. Infections amongst just two hidden populations (MSM, IDUs) accounted for the estimated 69% of 53,600 new HIV infections in the US during 2006. We propose to develop statistical methods for improving estimation in RDS samples with a focus on developing tools and methods that will help, and be accessible by, researchers in the field. From a statistical perspective, RDS is an adaptive sampling regime aimed at obtaining a probability sample. It is often effective at acquiring a sample, but the degree to which it can be considered a probability sample, with known inclusion probabilities, is unclear. The current estimators of these inclusion probabilities are known to be problematic. We will develop open-source user-friendly statistical software for RDS incorporating state-of-the-art methods, hold workshops to establish collaboration between applied and methodological researchers, disseminate these new methods and provide training using the tools developed. This project is significant because it will result in major advances in methodology for hidden population measurement and related scientific problems. The investigators are uniquely qualified as the first statisticians with close collaborative ties with field-researchers to systematize statistical understanding of RDS. The project is innovative in that it challenges the existing paradigm of RDS inference and proposes a new approach based on cutting edge statistical ideas and models. There is a dearth of statistical methodology justifying RDS. This project will produce a systematic statistical framework within which to understand the strengths and weaknesses of RDS. In the proposed work we will represent the complexities of the RDS procedure, but also allow the uncertainty of the resulting inference to be quantified. The development of statistical methodology for RDS is of vital importance to the social and behavioral sciences. We will disseminate the methodology for estimation, diagnostics and quantification of uncertainty via open-source user-friendly software aimed at field researchers. These will be applicable to both future RDS data, and the large existing data bases of RDS surveys. PUBLIC HEALTH RELEVANCE: Respondent Driven Sampling (RDS) is arguably the best and most common method being used to survey "hidden" populations that are hard to sample using standard probability methods. Examples include injection drug users (IDU), men who have sex with men (MSM), and female sex workers. Infections amongst just two hidden populations (MSM, IDUs) accounted for the estimated 69% of 53,600 new HIV infections in the US during 2006 (CDC report, Hall, 2008). The purpose of this grant is to improve these kinds of estimates as current estimation practices are known to be problematic.