What is Respondent-driven sampling?
Respondent-driven sampling (RDS) is a chain-referral survey method used to reach hard-to-reach or networked populations by having participants recruit peers. It combines peer recruitment with statistical weighting to allow limited population estimates from non-random samples.
Respondent-driven sampling starts with a small set of initial participants called seeds.
Each seed completes the survey and is given a limited number of recruitment coupons to invite peers from their social network. Recruited participants then do the same, producing recruitment chains or waves. Alongside responses, the survey collects information about each participant's network size (how many people they could have recruited). That network data is used with RDS estimators to adjust for the fact that people with larger networks are more likely to be recruited, which helps produce population-level estimates despite the non-random sampling. RDS is commonly used for populations that are difficult to sample by conventional means (for example, groups with no sampling frame) and requires careful design, monitoring and analysis to reduce bias.
Usage example
A health outreach team wants data from recent migrants in a city where there is no registry. They recruit 8 initial seeds from different communities, ask each to refer up to 3 peers, collect each participant's reported number of eligible peers, and use RDS weighting to estimate service uptake across the wider migrant population.
Practical application
RDS matters when you need information from communities that are hidden, dispersed, or distrustful of authorities and therefore unlikely to appear in standard sampling frames. It leverages existing social connections to increase participation and can produce approximate population estimates when random sampling isn't possible. For multilingual surveys, using RDS means designing recruitment materials and instructions in participants' preferred languages, ensuring translated consent and coupons, and tracking recruitment chains across language groups. That reduces barriers to participation and helps avoid excluding parts of a community simply because of language.
FAQ
How is respondent-driven sampling different from snowball sampling?
Both use peer referral, but RDS is more structured: it limits the number of recruits per participant, records recruitment chains and participants' network sizes, and applies statistical weighting to reduce bias. Snowball sampling is typically exploratory and does not include the same formal adjustment methods.
Can RDS produce representative estimates of a population?
RDS can produce useful population estimates when assumptions (e.g., sufficient recruitment waves, accurate reporting of network size, well-connected networks) are reasonably met and appropriate estimators are used. However, bias can remain—especially with strong homophily, isolated subgroups, or inaccurate network-size reports—so results should be interpreted with caution.
What are key practical tips for running RDS in multilingual communities?
Choose diverse seeds across language groups, translate recruitment coupons and instructions accurately, collect participants' language preferences, monitor recruitment chains to spot bottlenecks by language or community, and provide translated consent and help so recruits understand the process. Use a single survey that supports multiple languages and captures network-size consistently across translations.
Are there ethical or logistical concerns with respondent-driven sampling?
Yes. Incentives can pressure people to recruit inappropriately, privacy risks arise when people refer peers, and some communities may be harmed by visible recruitment. Design incentives, referral limits and confidentiality protections carefully, secure informed consent in participants' languages, and consult community stakeholders before using RDS.