Snowball Sampling — Chain-Referral Sampling
Snowball Sampling (Chain-Referral Sampling) · Also known as: chain-referral sampling, network sampling, respondent-driven sampling, referral sampling
Snowball sampling is a non-probability recruitment technique in which initial participants (seeds) refer the researcher to others who meet the study criteria, and those referrals in turn refer further participants. The sample grows incrementally — like a rolling snowball — until the required size or theoretical saturation is reached. It is the method of choice when a target population has no accessible sampling frame, such as undocumented migrants, illicit drug users, survivors of stigmatised experiences, or members of closed professional networks.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
+9 more
When to use it
Use snowball sampling when the target population is hidden, stigmatised, or lacks a sampling frame — such as people engaged in illegal behaviour, members of rare clinical groups, informal economy workers, or members of closed professional or social communities. It is also appropriate for exploratory qualitative research where the boundaries of a population are themselves unknown. Do NOT use snowball sampling when random selection is needed to support statistical inference about a defined population: because inclusion probability is unknown and correlated with social network membership, standard inferential statistics cannot be validly applied. Avoid it when the research question requires diverse coverage of the population and seeds come from a narrow social cluster, as the resulting sample will reproduce that cluster's homogeneity rather than capturing population variation.
Strengths & limitations
- Provides access to hidden, marginalised, or hard-to-reach populations for which no sampling frame exists.
- Leverages existing trust relationships, improving participant willingness to engage with sensitive topics.
- Low recruitment cost relative to other methods for locating rare or hidden sub-groups.
- Flexible and adaptable — chain length and branching factor can be adjusted as field conditions evolve.
- Compatible with both qualitative depth (interviews) and quantitative breadth (surveys), including respondent-driven sampling extensions.
- Non-probability method: inclusion probabilities are unknown, making statistical generalisation to the full population unjustifiable without special adjustments (e.g., respondent-driven sampling estimators).
- Sample is bounded by the social networks of seeds — individuals outside those networks have zero probability of inclusion, producing systematic under-coverage.
- Homophily bias: because people tend to refer others similar to themselves, the sample may over-represent certain sub-groups while missing others.
- Researcher has little control over sample composition; diversity depends on seed selection and referral chains, not design.
Frequently asked
Can I generalise findings from a snowball sample to the broader population?
Not with standard inferential statistics. Because inclusion probabilities are unknown and correlated with social network position, the sample is biased toward the seeds' networks. If population-level estimates are the goal, consider respondent-driven sampling, which uses chain-referral recruitment but attaches probability weights derived from network degree to enable valid estimation.
How many seeds should I start with?
The literature recommends beginning with multiple diverse seeds — typically three to six — drawn from different entry points into the target population. A single seed produces a narrow, homogeneous chain. More diverse seeds widen the network reach and reduce the risk that the sample reflects only one social cluster.
What is the difference between snowball sampling and respondent-driven sampling?
Respondent-driven sampling (RDS) is an extension of snowball sampling developed by Heckathorn (1997) that adds two features: participants are given a fixed number of uniquely coded coupons to distribute to peers, and dual-incentive structures reward both participation and successful referral. These features enable estimation of network degree and, in turn, allow calculation of inclusion probability weights that support population-level inference — something plain snowball sampling cannot offer.
When should I stop recruiting?
In qualitative designs, stop when thematic saturation is reached — when new interviews are not generating new themes, concepts, or categories. In quantitative designs, stop when the pre-specified sample size is reached. Track cumulative diversity across waves; if later waves merely repeat earlier findings, additional recruitment is unlikely to add value.
Is snowball sampling appropriate for mixed-methods research?
Yes, and it is frequently used in mixed-methods designs where the qualitative strand explores experience in depth and the quantitative strand documents prevalence or correlates within a hard-to-reach group. In these cases the limitations of the non-probability sampling frame should be acknowledged explicitly when reporting quantitative results.
Sources
- Goodman, L. A. (1961). Snowball sampling. Annals of Mathematical Statistics, 32(1), 148–170. DOI: 10.1214/aoms/1177705148 ↗
- Biernacki, P., & Waldorf, D. (1981). Snowball sampling: Problems and techniques of chain referral sampling. Sociological Methods and Research, 10(2), 141–163. DOI: 10.1177/004912418101000205 ↗
How to cite this page
ScholarGate. (2026, June 3). Snowball Sampling (Chain-Referral Sampling). ScholarGate. https://scholargate.app/en/survey-methodology/snowball-sampling
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Maximum Variation SamplingSurvey Methodology↔ compare
- Purposive samplingSurvey Methodology↔ compare
- Respondent-Driven SamplingSurvey Methodology↔ compare