Improving Reproducibility of Respondent Driven Sampling through Adaptive Design

受访者驱动抽样（RDS）是一种招募方法，用于难以抽样的人群， 由于高度污名化或非法行为，数量稀少和/或难以捉摸。对于这些群体来说，传统 概率抽样很少提供可行性，因为它需要过高的筛选成本来定位 符合条件的人，即使找到了符合条件的人，他们隐藏的愿望也会产生假阴性。 基于这样的前提，即具有相似特征的人形成某种类型的社交网络，RDS利用现有的 网络招聘，并已应用于许多研究。什么使RDS与传统 抽样的另一个特点是，招募过程主要由参与者自己通过他们的链条控制- 要求参与者从其网络中招募其他合格人员的推荐。使用有机社会 用于采样的网络是RDS的一个创新功能。然而，这带来了一个重大挑战。在 为了利用RDS，参与者需要配合招聘请求。因为 如果不合作，样本可能会停止增长，导致项目超支。但缺乏 在文献中对这种不合作过程的关注使得RDS数据收集过程变得极其困难 在设计阶段预测，当面临不期望的（通常是意想不到的）挑战时， 研究人员被迫进行计划外的设计改变（例如，提供更大的激励措施）， 希望能让RDS“发挥作用”。此外，不合作导致违反一个关键的 RDS推理的假设。总而言之，目前的RDS实践缺乏可操作性和统计性 可重复性，使其科学完整性受到质疑。 本研究试图通过提出自适应RDS（A-RDS）设计来提高RDS的重现性 框架，并提供实用的工具，研究人员依赖于成功实施的RDS和 通过开发A-RDS特定的设计指南和软件，可以监控RDS数据收集 进步和改进推理，密切反映真实的数据生成过程。根据A-RDS，我们将 计划设计适应战略，包括数据收集前的适应指标和规则。 在实地工作中，我们没有假设参与者之间的招聘合作模式相同， 将预测个人层面的合作倾向，从传入的数据和量身定制的数量和类型， 基于预先指定的规则为每个参与者接收优惠券。为此，数据收集工作 将被密切监测并用于作出适应决定。特别是，这种方法是经验性的。 应用于PWID研究，为解决阿片类药物使用迅速升级的问题提供数据。 通过为研究界提供一个实用的、数据驱动的、基于规则的工具，拟议的研究将 提高研究人员对RDS操作的控制，不仅提高了重现性， 满足RDS中有效推理所需的关键假设的机会增加。

Respondent driven sampling (RDS) is a recruitment method for hard-to-sample populations that are rare in number and/or elusive due to highly-stigmatized or illicit behaviors. For these groups, traditional probability sampling rarely offers feasibility, because it requires prohibitively high screening costs to locate eligible persons, and, even when eligible persons are located, their desire to hide produces false negatives. Based on the premise that people of similar traits form some type of social networks, RDS exploits the existing networks for recruitment and has been applied to numerous studies. What sets RDS apart from traditional sampling is that the recruitment process is mostly controlled by participants themselves through their chain- referral that asks participants to recruit other eligible persons from their networks. The use of organic social networks for sampling is an innovative feature of RDS. This, however, comes with one major challenge. In order to capitalize on RDS, participants need to cooperate with recruitment requests. Because of noncooperation, the sample may stop growing in size, resulting in a project overrun. However, the lack of attention to this noncooperation process in the literature makes RDS data collection progress extremely difficult to predict at the design stage, and when faces with undesirable (and often unexpected) challenges, researchers are forced to make unplanned design changes (e.g., offering larger incentives) on the spur of the moment in hopes of making RDS “work”. Additionally, noncooperation leads to a violation of a critical assumption of RDS inferences. In sum, the current practice of RDS lacks operational and statistical reproducibility, making its scientific integrity questionable. This study attempts to improve reproducibility of RDS by proposing Adaptive-RDS (A-RDS) as a design framework and to provide practical tools on which researchers rely for successful implementation of RDS and by developing A-RDS specific design guidelines and software that will allow monitoring RDS data collection progress and improve inferences that closely mirror the true data generation process. Under A-RDS, we will plan design adaptation strategies, including indicators and rules for adaptations prior to the data collection. During the field work, instead of assuming the same recruitment cooperation patterns across participants, we will predict individual-level cooperation propensities from incoming data and tailor the number and type of coupons for each participant received based on the pre-specified rules. For doing so, data collection progress will be closely monitored and used for making adaptation decisions. In particular, this approach is empirically applied to PWID studies to provide data for addressing rapidly escalated issues with opioid use. By providing a practical yet data-driven, rule-based tool to the research community, the proposed study will boost researchers' control on the operations of RDS, leading to not only improved reproducibility but also increased chances of meeting critical assumptions in RDS required for valid inferences.