A chain‑referral sampling method for hard‑to‑reach or networked populations in which selected seeds recruit peers using a limited number of coupons, recruitment chains are tracked, and analysts collect network degree and recruitment information to apply estimators that weight observations to approximate population parameters under specific network and recruitment assumptions.

Principle

Peer recruitment through social ties permits access to otherwise inaccessible populations; by recording recruitment patterns and respondents’ personal network sizes (degree) and applying model‑based weights, analysts can attempt to correct for differential inclusion probabilities—provided key assumptions (sufficient waves, random recruitment conditional on degree, accurate degree reporting, and a single connected network component) are approximately met.

Demonstration

Illustrative scenario → A study of a stigmatized urban population begins with 8 purposive seeds who each receive 3 coupons to recruit peers. Recruitment proceeds for several waves; the study records who recruited whom and asks each respondent about the number of eligible peers they know. Analysts use recruitment trees, reported degree, and appropriate RDS estimators to produce weighted prevalence estimates for behaviors of interest and compute design‑based confidence intervals acknowledging network dependence.

Misapplication

Treating RDS data as a simple random sample and ignoring network dependence and the required weighting (error: underestimating bias and variance), or failing to record or use degree and recruitment information when applying estimators—both practices yield biased point estimates and invalid uncertainty measures.

Consequence

When assumptions are at least approximately satisfied and data are properly recorded and weighted, RDS can produce useful population estimates where probability sampling is infeasible; when assumptions fail (e.g., extreme homophily, inaccurate degree reporting, disconnected subnets), estimates may be substantially biased and uncertainty understated.

Reversal

If the underlying social network is disconnected, recruitment chains remain short, respondents systematically misreport degree, or recruitment is highly non‑random within degree strata, the weighting model breaks down and RDS estimators become unreliable; in such contexts alternative approaches (multiple venue sampling, targeted probability methods) may be preferable.

Boundary

Clearly within: studies of hidden, networked populations where peer referral is feasible and seeds can initiate long recruitment chains. Boundary case: partially networked populations or studies with very few waves or small sample sizes (estimator assumptions tenuous). Clearly outside: non‑network convenience samples, single‑stage venue sampling without controlled peer recruitment, or chain referral without recorded recruitment and degree data (classic snowball sampling).

Semantic Tension

Practical access and feasibility (ability to recruit otherwise unreachable respondents) versus strict statistical assumptions required for unbiased estimation; researchers must trade off field feasibility against the credibility of inferential claims.

Synthesis

RDS is a pragmatic compromise: it uses controlled peer recruitment and degree‑based weighting to convert chain‑referral data into population estimates under explicit network assumptions. Its validity depends critically on meeting those assumptions and on transparent reporting of recruitment dynamics and diagnostic checks.