Abstract
<sec> <title>UNSTRUCTURED</title> <p>Respondent-driven sampling (RDS) is widely used to recruit hard-to-sample populations, including racial and ethnic minority groups, yet limited evidence exists on how seed selection mechanisms affect data quality. This study examines seed selection in the Health and Well-Being of Koreans (HAWK) study, a national web-based RDS survey of Korean American adults. We experimentally varied seed recruitment using two mechanisms: a probability-oriented postal address list associated with Korean ethnicity or surnames (“mail seeds”) and convenience-based online recruitment through platforms such as Facebook (“non-mail seeds”). The final sample included 772 Korean American adults from 57 seeds. We evaluated geographic coverage and benchmarked HAWK estimates against 2022 American Community Survey (ACS) estimates for 18 demographic, socioeconomic, immigration, health care access, and disability variables. We also assessed whether statistical adjustment methods improved the quality of HAWK estimates. RDS expanded the geographic coverage of the sample, from seeds located in 25 states to respondents in 36 states. However, data quality varied by seed selection mechanism. Samples generated from mail seeds more closely approximated ACS benchmarks than samples generated from non-mail seeds, particularly for age, education, nativity, citizenship, and immigration-related measures. Statistical adjustments generally increased standard errors without improving bias reduction. Findings suggest that Web-RDS can effectively reach geographically dispersed racial and ethnic minority populations, but that seed selection has nontrivial consequences for data quality. Probability-oriented seed recruitment may improve data quality more effectively than statistical adjustments. Future RDS studies should give greater attention to seed recruitment mechanisms.</p> </sec>