Home Statistical Dictionary Respondent-Driven Sampling
Respondent-Driven Sampling
A structured refinement of snowball sampling that tracks recruitment chains and each participant's personal network size, using that information to construct statistical weights intended to approximate a probability sample of a hidden population.
In Plain English
Respondent-driven sampling (RDS) takes the basic idea of snowball sampling, participants recruit other participants, and adds mathematical structure on top of it: participants are given a limited number of numbered referral coupons, recruitment chains are carefully tracked wave by wave, and each person is asked how large their personal network is (how many people they know who'd be eligible for the study). That extra bookkeeping lets researchers build statistical weights that attempt to correct for the fact that well-connected people are more likely to get recruited, turning what would otherwise be an uncontrolled snowball into something researchers can analyze with formal, if still debated, estimation methods, making it one of the standard tools for studying hidden populations like people who inject drugs or sex workers where no ordinary sampling frame could ever exist.
Definition
Respondent-driven sampling (RDS) is a peer-referral sampling method, structurally a controlled variant of snowball sampling, in which recruitment proceeds through tracked chains starting from a small set of seeds, each participant receives a limited number of uniquely coded referral coupons to recruit peers, and reports their personal network size (degree), the number of people in the target population they know. Under a Markov-chain model of the recruitment process, RDS estimators, most notably the Volz-Heckathorn estimator, use each participant's reported network size to construct an inverse-degree weight, participants with smaller personal networks are up-weighted and those with larger networks are down-weighted, on the reasoning that a person with more connections had a proportionally higher chance of being recruited into the sample, an approach intended to asymptotically approximate a probability sample of the hidden population as the number of recruitment waves grows, though the validity of this asymptotic approximation, and its sensitivity to network structure and non-random referral behavior, remains a topic of active methodological research and debate.
Formula
Notation
Properties
- RDS's central statistical claim, that inverse-degree weighting corrects for the recruitment advantage of well-connected participants and produces asymptotically unbiased population estimates, depends on strong assumptions about the underlying social network (such as recruitment proportional to degree and a well-mixed network), assumptions that are difficult to verify from the sample data itself and are an active area of methodological critique.
- The coupon system, giving each participant a small, fixed number of numbered referral coupons rather than unlimited referral, is a deliberate design feature intended to prevent any single highly connected 'super-recruiter' from dominating the sample and to allow researchers to reconstruct the exact recruitment chain for later analysis.
- RDS is one of the primary methods recommended by public health agencies (including for HIV surveillance in key populations) specifically because it targets populations, such as people who inject drugs, sex workers, or men who have sex with men in criminalized contexts, for which no conventional sampling frame could ever be constructed, making some structured alternative to unmonitored snowball sampling a practical necessity.
At a Glance