Injection Drug User (IDU) Peer Recruitment Dynamics and Network Structure in Respondent Driven Sampling (RDS)
Respondent Driven Sampling (RDS) has gained its popularity internationally for its cost effectiveness in reaching hidden populations and claimed ability to make unbiased population estimates. However, RDS inference models were based on strong assumptions regarding peer recruitment processes and the structure of underlying social networks, which have not yet been empirically scrutinized. We conducted a mixed method intensive social network study in Hartford, CT and recruit a typical RDS sample of 524 injection drug users (IDUs). Comprehensive social network surveys before and after actual peer recruitment, location information, as well as 60 in-depth interviews will be analyzed to assess IDUs’ peer recruitment intentions, actual experiences, and real world contextual factors associated with coupon distribution and return success. This study is the first attempt to scrutinize an RDS sample as complex multiple-layered networks linked by different social ties specifically related to RDS sample recruitment progresses using mixed methods. Findings will have direct application to development of improved RDS estimators; or to assess performance of existing estimators. This Study addresses the challenges faced by epidemiologists and policy makers to better understand the HIV risk profile among injection drug using populations and other hidden high-risk groups.
JiangHong Li, M.D., M.S.
(860) 278-2044 ext. 297
Margaret R. Weeks
Gayatri Moorthi, Ph.D.
Chiekwu Obidoa, Ph.D.
Heather Mosher, Ph.D.
University of Southern California
Thomas Valente, Ph.D.
Robert Heimer, Ph.D.
With increasing application of RDS among a variety of populations in different contexts, serious skepticism has arisen regarding the validity of RDS statistical inference models. HIV researchers and epidemiologists have reported challenges in meeting the underlying assumptions during implementations and question the implications for the validity of population estimates (4, 11-14). The most widely used S-H estimator (10), even under ideal implementation situations, was found to produce inferences substantially less accurate than was generally acknowledged, although unbiased (15-17). RDS statisticians have been developing new estimators over the past 10 years (10, 18-21) based on more realistic assumptions about participants’ social network structure and peer recruitment process and dynamics (17). However, as a network-based special snowball sampling method, all RDS inference models, including the recently improved estimators, were still based on strong assumptions. These include: respondents’ self-reported degree (network size) is accurate; recruitment occurs over the recruiter’s network ties (i.e., directly to personal network members); selection of recruits in random (also referred as “uniform recruitment/selection”, in contrast to “preferential recruitment”); and recruit’s enrollment probability is in proportion to his/her degree. However, none of these assumptions has been systematically and empirically tested due to the challenge of observing hidden population networks.
As new RDS inference models are loosening up strong assumptions and building up more realistic assumptions about population network structures and peer recruitment behaviors, the inference accuracy also improves (17, 18, 22). This points to the importance of understanding the real world implementation processed of peer recruitment dynamics. There is a critical need for empirical understanding of peer recruitment behavior and processes, recruits’ response patterns, the nature of social network linkages and network changes as RDS is implemented in real world settings. Such knowledge will have implications for future development of improved RDS estimators, RDS assumption violation diagnostic tools, and better designs of survey questions to estimate network degree. The specific aims of the study are:
1) Recruit a sample of IDUs using RDS and simultaneously conduct a social network study of recruited individuals.
2) Understand factors that influence peer recruitment intention decision making, dynamics of recruitment attempts, enrollment attrition and changes in influences over time as peer recruitment proceeds.
3) Understand the composition and structures of IDUs’ multi-layered social networks (i.e., the infection risk network, the intent and actual peer recruitment network, and final enrollment network members), and the associations among them.
The RDS-Net study is the first attempt to scrutinize an RDS sample as complex multiple-layered network linked by different social ties specifically related to RDS sample recruitment progresses using mixed methods. Findings from this study will have direct application to development of improved RDS estimators or to assess performance of existing estimators needed to improve population risk estimates.
Among 408 non-seed participants who returned to the 2-month follow-up, only 49.6% reported recruiter name match with the original coupon holder, 41.8% were non-matched, and 8.5% were unclear. Unexpectedly, nearly half of the coupons were redistributed on the street before arriving at the final recruit enrolled in the study. These results failed the most important RDS estimator assumptions on which all RDS models were based, that recruitment occurs over the recruiter’s social network ties (10, 18, 46-48). It also challenges the foundation of earlier Heckathorn, Salganik-Heckathorn estimators (7, 10) which heavily rely on accuracy of the coupon-suggested recruiter-recruit relationship.
Intensive social network data collection and construction of a sociometric network enabled us, for the first time in RDS research, to uncover unobserved network nodes from a smaller number of sampled hard to reach urban IDUs. This uncovered sociometric network dataset, along with comprehensive survey measures, captured nuanced peer recruitment processes and dynamics at the individual level, dyadic relationship level and whole network level. These data revealed several significant yet overlooked threats to previous RDS estimators. Several most common assumptions about peer recruitment behavior and processes failed the assessment. These include that peer recruitment occurs over the recruiters’ network ties, random selection of recruits included in sample, and recruits’ probability of being in the sample is proportional to their degree. The source of most problems is rooted in the overly simplistic estimator validity assessments (15, 49, 50) failed to recognize the complexity of the peer recruitment process, which involved multiple people, such as the initial recruiter, their initial targeted recruit candidates, alternative candidates, coupon receivers, their network members, final coupon receivers, and staff. These people’s decisions, behaviors, and interactions with others occur along a time horizon that starts with the recruiter planning, trying to reach recruit candidates, interacting with the candidates, passing the coupon to them, or seek other recruit candidates to pass on the coupon or let it expire. After a coupon is successfully passed, what happens is beyond the initial recruiter’s control. The initial coupon receiver could either overcome various barriers and enter the study, let coupon expire, or pass it on to others. Once the latter happens, the final successful recruit, an indirect recruit, may not be in the initial recruiter’s network. Our data show that nearly half of the sample were indirectly recruited. Thus, failure of the assumption that “peer recruitment occurs over the recruiter’s network ties” is perhaps the greatest threat to all RDS estimators’ validity.