Exploring Design Aspects of Web-Based Respondent-Driven Sampling for Racial/Ethnic Minorities

摘要 增加少数民族数据的可用性是HHS减少种族和族裔健康行动计划的优先事项 差异。然而，目前关于少数种族/族裔的研究数据不足以说明其 样本量，因此不能提供足够的统计功效进行分析。基于Web的响应驱动 Web-RDS是RDS的扩展，它利用现有的社交网络进行招聘研究 参与者，特别是罕见的，隐藏的和/或难以接触的群体。而不是研究人员招募 在RDS中，参与者从自己的社交网络中招募其他合资格的人。这 招聘过程仍在继续，主要是通过激励招聘券进行。因此，我们认为， RDS有潜力捕捉那些研究人员难以接触到的人，而且成本要低得多 比传统的抽样方法更严格。虽然RDS通常是亲自管理，但Web-RDS适用于 RDS on the Web，它消除了对访问者的需求以及与 时间和地理。一旦建立了数据收集系统，Web-RDS的边际成本是 远低于RDS，进一步降低成本负担。与此同时，种族/族裔少数群体 据报道，他们形成了紧密的内部社交网络，并以非常高的速度访问网络，大约90%， 与非少数民族几乎相同。强大的社交网络与高Web访问率相结合 少数群体之间的联系使Web-RDS成为少数群体数据收集的一个有吸引力的平台。但有一个 关于铁路发展系统设计方面的文献中存在明显空白。设计问题不仅是实际问题， 这是任何Web-RDS研究都可能寻求的答案，也是影响数据质量的关键因素。 本研究试图填补这一空白，提供实用的设计指导方针和工具，为网络RDS的目标， 通过成功的实施提高数据质量，其中成功是用示例来衡量的 组成和招募倾向。具体来说，我们将探讨两个具体的设计元素（seed 选择和试样设计）通过随机化实验，并开发一个数据收集监测 系统为了做到这一点，Web-RDS将应用于对韩裔美国人的全国调查，这是一个罕见的少数民族 这一群体占美国人口的比例不到1%。尽管罕见，但基本的社会人口统计学 许多种族/少数民族群体，如韩裔美国人，可从美国 社区调查。这是特别有利的，因为不同环境下的Web-RDS数据的质量不同。 实验条件可以根据少数群体特定的ACS数据进行验证。通过经验证明Web- 通过设计性试验，将RDS作为少数群体数据收集方法及其数据质量， 拟议的研究将把这些实际的RDS具体设计问题的文献，并增加 认识到设计问题对数据质量的重要性。

ABSTRACT Increasing minority data availability is a priority of the HHS Action Plan to Reduce Racial and Ethnic Health Disparities. Currently available research data for racial/ethnicity minorities are, however, inadequate in their sample sizes and, hence, do not provide sufficient statistical power for analysis. Web-based respondent driven sampling (Web-RDS) is an extension of RDS, which exploits existing social networks for recruiting research participants, specifically, of rare, hidden and/or hard-to-reach groups. Rather than researchers recruiting participants directly, in RDS, participants recruit other eligible persons from own social networks. This recruitment process continues in waves and is mostly handled by incentivized recruitment coupons. Hence, RDS has potential to capture those who, otherwise, are difficult for researchers to reach and is much less cost intensive than traditional sampling methods. While RDS is typically administered in person, Web-RDS applies RDS on the Web, which leads to eliminating the need for interviewers and the constraints associated with the time and geography. Once the data collection system is established, marginal costs for Web-RDS is considerably lower than RDS, further reducing cost burdens. At the same time, racial/ethnicity minorities are reported to form tight ingroup social networks and to access the Web at a remarkably high rate, around 90%, a figure virtually the same as non-minorities'. The strong social network combined with the high Web access rate among minorities makes Web-RDS an attractive platform for minority data collection. However, there is a notable void in the literature on design aspects of RDS. Design issues are not only practical questions for which any Web-RDS studies are likely to seek answers but also a critical factor influencing data quality. This study attempts to fill this gap by providing practical design guidelines and tools for Web-RDS for a goal of improving data quality through successful implementations, where success is measured with sample composition and recruitment propensity. Specifically, we will explore two specific design elements (seed selection and coupon design) through randomized experiments and develop a data collection monitoring system. For doing so, Web-RDS will be applied to a national survey of Korean Americans, a rare minority group that comprises less than 1% of the U.S. population. Despite being rare, basic socio-demographic information is available for many racial/ethnic minority groups, such as Korean Americans, from the American Community Survey. This is particularly advantageous as the quality of the Web-RDS data under different experimental conditions can be verified against the minority-specific ACS data. By empirically demonstrating Web- RDS as a method for minority-specific data collection and its data quality through design experiments, the proposed study will bring these practical RDS-specific design issues to the literature and increase the awareness about the importance of design issues for data quality.