Respondent-Driven Sampling
Also known as: Chain-Referral Sampling, Peer-Referral Sampling, Network-Based Sampling, Katılımcı Güdümlü Örnekleme
Respondent-Driven Sampling (RDS) is a probabilistic chain-referral method designed to reach hidden or hard-to-reach populations that lack a sampling frame. Introduced by sociologist Douglas Heckathorn in 1997, RDS combines snowball recruitment with mathematical weighting based on participants' personal network sizes, allowing researchers to generate population-level estimates even when no complete membership list exists.
Key highlights
- Enables probabilistic inference for hidden populations with no sampling frame
- Inverse-degree weighting corrects for differential recruitment probability
- Markov chain convergence ensures estimates are independent of seed selection after sufficient waves
- Dual incentive structure (participation + recruitment) boosts response and referral rates
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
RDS is appropriate when the target population is hidden, stigmatized, or lacks a sampling frame — such as people who inject drugs, sex workers, undocumented migrants, or men who have sex with men. Key assumptions include: the population forms a connected network, participants recruit randomly from their contacts, and network degree is reported accurately. RDS performs poorly in fragmented networks or when incentives create biased referrals. Alternatives include time-location sampling and capture-recapture methods.
Strengths & limitations
- Enables probabilistic inference for hidden populations with no sampling frame
- Inverse-degree weighting corrects for differential recruitment probability
- Markov chain convergence ensures estimates are independent of seed selection after sufficient waves
- Dual incentive structure (participation + recruitment) boosts response and referral rates
- Assumes a single connected network component; fragmented populations violate convergence conditions
- Relies on accurate self-reported network degree, which is subject to recall and social desirability bias
- Variance estimation is complex and standard errors are typically larger than in simple random samples
- Long recruitment chains can introduce homophily bias if peers are disproportionately similar to recruiters
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How many seeds are needed to start an RDS study?
Most implementations use 5 to 10 seeds representing different subgroups of the target population. The exact number matters less than diversity: because the Markov chain eventually converges to a stationary distribution, estimates from later waves become insensitive to seed choices, provided the network is well-connected and enough waves are completed.
What software is available for RDS analysis?
RDS Analyst (formerly RDSAT) is the most widely used dedicated tool; it implements multiple estimators including the Heckathorn and Volz-Heckathorn approaches. The R package 'RDS' provides similar functionality along with diagnostics for convergence, homophily, and bottleneck detection, making it suitable for reproducible research workflows.
How does RDS differ from snowball sampling?
Snowball sampling is a convenience technique with no formal probabilistic basis, making population-level inference impossible. RDS adds two elements that enable inference: (1) strict coupon tracking that records the referral network, and (2) inverse-degree weighting that corrects for unequal selection probabilities. Together, these allow RDS estimates to approximate those from a probability sample.
Sources
- 1.Heckathorn, D. D. (1997). Respondent-driven sampling: A new approach to the study of hidden populations. Social Problems, 44(2), 174–199.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 2). Respondent-Driven Sampling. ScholarGate. https://scholargate.app/survey-methodology/respondent-driven-sampling