Snowball Sampling vs. Respondent-Driven Sampling: What’s the Difference?

Are Snowball Sampling and Respondent-Driven Sampling Just Two Names for Peer Recruitment?

Suppose you are investigating the experiences of independent researchers who work outside universities and formal research institutions.

You cannot obtain a complete list of eligible individuals, so you begin by contacting several researchers and asking them to recommend others who might participate.

Your recruitment gradually expands through referrals. You describe the approach as snowball sampling.

A colleague suggests that the method should instead be called respondent-driven sampling because participants are recruiting other participants.

Would changing the label make the sample more representative? Does respondent-driven sampling require additional procedures, and can either method support estimates about the entire population?

Although both methods use social connections to identify participants, their methodological purposes and inferential assumptions differ substantially.

Both Use Peer Referrals, but RDS Adds a Structured Recruitment and Estimation Framework

How Snowball Sampling and Respondent-Driven Sampling Differ

What is snowball sampling?

Snowball sampling is a recruitment approach in which researchers use existing contacts or participants to identify additional individuals who may satisfy the study's eligibility criteria.

For example, a researcher studying independent scholars may interview several eligible individuals and ask whether they know others with relevant experiences.

The newly identified individuals may then provide further referrals.

Recruitment expands through social relationships, often reaching people who are absent from institutional directories or conventional sampling frames.

Snowball sampling is commonly classified as a nonprobability sampling approach because the inclusion probabilities of population members are generally unknown.

However, the term describes a family of referral practices rather than one universally standardized procedure.

Researchers should specify how referrals are requested, how participants are selected, and whether recruitment is guided by particular characteristics or emerging information needs.

What is respondent-driven sampling?

Respondent-driven sampling, or RDS, is a structured network-based recruitment method developed to study populations that are difficult to identify through conventional sampling frames.

It typically begins with selected seeds who receive a limited number of recruitment invitations for eligible peers.

Participants recruited through those invitations may recruit additional peers, creating tracked recruitment chains and successive waves.

RDS studies commonly collect information about participants' eligible social network sizes and use specialized estimators when population inference is intended.

However, the validity of such estimates depends on the assumptions of the particular estimator and recruitment process.

RDS should therefore be understood as a structured recruitment and inferential framework rather than simply a more sophisticated name for snowball sampling.

What are the main differences?

These distinctions describe common implementations. Individual studies may use variations, and researchers should report the actual procedures rather than rely exclusively on the sampling label.

Why are both methods called chain-referral approaches?

Both methods use existing social connections to extend recruitment beyond the individuals initially identified by researchers.

For example, Participant A may refer Participant B, who then refers Participant C.

This creates a recruitment chain.

However, chain referral describes the mechanism for reaching participants, not the statistical properties of the resulting sample.

Different referral procedures can produce different patterns of selection.

Researchers should therefore avoid assuming that all chain-referral methods have the same inferential capabilities.

How does the researcher's role in recruitment differ?

In snowball sampling, researchers may retain substantial discretion over which referrals to pursue.

For example, they may prioritize individuals with contrasting experiences, specialized knowledge, or characteristics not yet represented in the sample.

This can be appropriate in qualitative research where purposive variation or information-rich cases are important.

In RDS, recruitment is more formally organized through participant-to-participant referrals under specified rules.

Researchers commonly track the referral process and seek to limit uncontrolled recruitment decisions.

However, participants still exercise discretion over whom they invite, and that discretion may influence selection.

The distinction is therefore one of structured recruitment mechanisms, not complete elimination of human selection.

What is the difference between a referral and a recruitment coupon?

In ordinary snowball sampling, a referral may involve sharing study information, providing contact details with appropriate permission, or introducing the researcher to another eligible individual.

In RDS, a recruitment coupon or controlled invitation typically identifies a referral opportunity and helps track the recruitment chain.

Coupons may be physical or digital, depending on the study.

They can help researchers document which participant recruited another participant and how recruitment waves develop.

However, distributing coupons does not automatically transform snowball sampling into valid RDS.

Network measurement, estimator assumptions, recruitment procedures, and the intended inferential objective must also be considered.

Why does RDS measure personal network size?

Some RDS estimators use the number of eligible social connections reported by each participant to adjust for unequal recruitment opportunities.

Individuals with more relevant connections may have more opportunities to receive invitations.

Under certain theoretical assumptions, inverse-degree weighting can help account for this difference.

However, the network question must be carefully defined.

For example, asking how many independent researchers a participant knows may produce a different answer from asking how many eligible independent researchers they know personally and could contact.

Inaccurate network-degree reports may undermine estimation.

Snowball sampling does not ordinarily require degree measurement because it generally does not claim to use degree-based population estimators.

Can snowball sampling produce representative population estimates?

Ordinary snowball sampling generally does not provide known inclusion probabilities for members of the target population.

Participants may be recruited because they are socially connected to initial contacts, willing to refer others, or visible within particular networks.

These processes may systematically exclude isolated individuals and underrepresent less connected groups.

Consequently, an observed sample proportion should not automatically be interpreted as a population prevalence estimate.

For example, if 60% of snowball-recruited independent researchers report a particular challenge, that does not establish that 60% of all independent researchers experience it.

Population inference would require additional assumptions or evidence not supplied by ordinary snowball recruitment alone.

Can RDS produce representative population estimates?

RDS was developed partly to support population estimation from structured network recruitment.

However, its estimators rely on assumptions about social networks, recruitment behavior, network-degree measurement, and other features of the sampling process.

For example, recruitment may be concentrated among people with similar characteristics, and some population groups may be disconnected from the sampled network.

Weighting cannot automatically correct these problems.

Researchers should therefore describe RDS estimates as conditional on the estimator and its assumptions rather than automatically representative.

The methodological requirements are discussed in when respondent-driven sampling is appropriate.

Why does social network structure matter more for RDS inference?

Social networks affect both methods because participants recruit through their relationships.

However, RDS population estimation explicitly depends on assumptions about how those relationships influence recruitment.

For example, a population may contain several communities with few connections between them.

Recruitment chains beginning in one community may not reach the others.

In ordinary snowball sampling, this limitation may be addressed by deliberately seeking additional contacts to improve diversity, particularly for qualitative objectives.

In RDS, network segregation can affect the validity of the estimator and must be considered in relation to the assumed recruitment process.

Neither approach automatically reaches people who are socially disconnected from the initial recruitment networks.

What is homophily, and how does it affect both methods?

Homophily is the tendency for individuals to form social relationships with others who share particular characteristics.

For example, independent researchers in the same discipline may be more likely to know one another than researchers working in unrelated fields.

Snowball recruitment may therefore concentrate on particular disciplinary communities.

RDS recruitment chains may similarly remain within subgroups, complicating network mixing and population estimation.

Researchers should consider whether referrals are likely to cross relevant social boundaries.

Adding more participants from the same network does not necessarily improve coverage of excluded groups.

Which method is more appropriate for qualitative research?

Snowball sampling is often more directly aligned with qualitative research when the objective is to identify participants with relevant experiences or perspectives.

For example, a researcher may use referrals to locate independent scholars who have experienced different forms of institutional exclusion.

The researcher may deliberately seek variation in discipline, career stage, or publication experience.

However, the sampling strategy should be justified according to the qualitative research question.

Referral recruitment alone does not establish that the sample contains sufficiently diverse or informative experiences.

RDS-style procedures can also be used for recruitment in qualitative studies, but specialized population estimation may be unnecessary when the objective does not involve prevalence or population parameters.

Which method is more appropriate for quantitative research?

Both methods can recruit participants for quantitative data collection, but they support different inferential claims.

Snowball sampling may be used for exploratory surveys or studies focused on describing recruited participants.

However, population prevalence estimates based on ordinary snowball samples require particular caution.

RDS may be considered when the target population is network-connected and the study aims to estimate population characteristics using an appropriate RDS estimator.

Yet the method's assumptions must be examined, and the resulting estimates may remain biased if those assumptions fail.

The choice should therefore follow the estimand and sampling mechanism rather than the assumption that quantitative research always requires RDS.

Can snowball sampling be used for hypothesis testing?

Statistical tests can be calculated from snowball-recruited data, but their interpretation depends on the sampling process, dependence structure, and model assumptions.

For example, participants recruited through shared networks may have correlated characteristics or outcomes.

Ordinary independent-observation methods may therefore underestimate uncertainty.

Furthermore, statistically significant associations within a selectively recruited sample do not automatically generalize to the target population.

Researchers should distinguish computational feasibility from valid population inference.

Using a familiar statistical test does not convert nonprobability recruitment into probability sampling.

How do sample size requirements differ?

Snowball sampling sample size planning should reflect the study's objective, recruitment feasibility, and intended analysis.

For qualitative research, relevant considerations include the research question, diversity of experiences, depth of evidence, and analytical approach.

For quantitative snowball samples, conventional power or precision calculations may not capture selection bias and network dependence.

RDS sample size planning should additionally consider recruitment chains, network structure, estimator performance, and possible design effects.

There is no universal RDS design effect or fixed number of recruitment waves that guarantees adequate inference.

Researchers should avoid assuming that a large sample resolves the methodological limitations of either approach.

How do incentives differ between the methods?

Snowball sampling may compensate participants for participation, but referral incentives are not an inherent feature of the method.

RDS studies commonly use structured incentives for participation and successful eligible referrals.

However, incentive arrangements vary, and dual incentives are not sufficient to define RDS.

Researchers should evaluate whether incentives create pressure, encourage inappropriate disclosure, or influence whom participants recruit.

Ethical compensation should remain proportionate and compatible with voluntary participation.

Referral incentives do not establish statistical representativeness.

What ethical concerns do both approaches share?

Both methods involve social relationships that may reveal information about potential participants.

For example, asking one participant to identify others with sensitive experiences could inadvertently disclose information those individuals have not consented to share.

Researchers should consider allowing participants to forward study invitations rather than providing third-party contact details without permission.

Recruitment should preserve voluntary participation and avoid coercion through personal or institutional relationships.

Researchers should also minimize unnecessary identifying information and establish appropriate data security procedures.

RDS recruitment tracking introduces additional confidentiality considerations because referral relationships may be recorded.

These concerns should be addressed during ethics review and protocol development.

Can digital recruitment be combined with either method?

Yes. Snowball referrals may occur through email, messaging applications, online communities, or social platforms.

RDS can also be adapted to digital environments using controlled invitations and recruitment tracking.

However, digital recruitment may exclude individuals with limited connectivity and may alter the social network process through which referrals occur.

Invitations can also be forwarded widely, making recruitment chains difficult to interpret.

Researchers should evaluate whether the digital implementation preserves the procedures and assumptions of the selected method.

See digital recruitment for hidden populations for related considerations.

What if researchers combine purposive and snowball sampling?

Combining purposive and snowball recruitment can be appropriate when the researcher needs particular participant characteristics and referrals help identify eligible individuals.

For example, researchers may purposively recruit independent scholars from different disciplines and then use referrals to locate others with relevant experiences.

The sampling strategy should explain both the initial selection criteria and how referrals contribute to the intended variation.

However, the combination remains a nonprobability approach unless additional design features establish a defensible probability mechanism.

Researchers should avoid relabeling such a procedure as RDS merely because recruitment occurs through several waves.

Can snowball sampling be converted into RDS after data collection?

Generally, researchers should not retrospectively relabel an ordinary snowball sample as RDS simply because referral information was collected.

RDS estimation requires particular recruitment procedures and information that should be planned before recruitment.

For example, the study may need controlled invitations, recruitment-chain records, and clearly defined eligible network-degree questions.

If these elements were absent, the data may not support the assumptions or estimators intended for RDS.

Researchers should describe the actual recruitment process rather than change the methodological label to imply stronger inferential properties.

What if recruitment chains stop early?

In snowball sampling, early chain termination may limit the diversity or depth of the achieved sample.

Researchers may seek additional initial contacts or adjust recruitment channels according to the study's sampling rationale.

In RDS, early termination may prevent sufficient recruitment waves or produce excessive dependence on initial seeds.

Researchers should examine whether the achieved network recruitment process supports the selected estimator.

Adding new seeds or modifying recruitment rules may be necessary for feasibility, but those changes should be documented and evaluated for their inferential consequences.

Recruitment failure should not be concealed by reporting only the final sample size.

How should the two methods be reported?

For snowball sampling, researchers should describe initial participant selection, referral procedures, eligibility screening, recruitment channels, and the intended sampling rationale.

They should explain how the referral process may have affected the characteristics of the achieved sample.

For RDS, reporting should additionally describe seed selection, recruitment limits, referral chains, recruitment waves, network-degree measurement, the selected estimator, and relevant diagnostics.

Population estimates should be distinguished from unweighted sample descriptions.

Both methods require transparent reporting of recruitment limitations and ethical safeguards.

The methods should be named according to the procedures actually implemented, not according to the perceived strength of the label.

How should researchers choose between them?

Begin with the research question and determine whether the study seeks detailed understanding of selected participants or statistical inference about a defined population.

If the primary goal is identifying information-rich cases through referrals, snowball sampling may be appropriate.

If the goal is population estimation and the target population has a suitable network structure, RDS may be considered.

However, RDS should not be selected unless the recruitment process and estimator assumptions are defensible.

If neither method can support the intended inference, researchers should consider alternative sampling frames, community partnerships, or a revised research objective.

The broader problem is addressed in research design for hard-to-reach populations.

Choosing a Sampling Strategy for Independent Researchers Outside Universities

The example demonstrates that the same population may justify different sampling approaches for different research objectives.

Study A does not require representative prevalence estimation to answer its qualitative question. Study B does require an inferential strategy capable of supporting population estimates, and RDS is appropriate only if the necessary conditions are sufficiently plausible.

Common Misconceptions About Snowball Sampling and RDS

Choose According to the Inferential Goal, Not the Perceived Sophistication of the Method

Begin by determining whether referrals are needed primarily to identify participants or to support a structured network-based population estimation design.

Then assess whether the population's social network and the available recruitment procedures support the intended approach.

The choice should be documented before recruitment wherever feasible, including the intended inferential scope and the limitations of the recruitment process.

Neither method should be evaluated solely by how many participants it produces. The more important issue is whether the resulting evidence can answer the research question credibly.

Before Choosing Snowball Sampling or Respondent-Driven Sampling

Frequently Asked Questions About Snowball Sampling and RDS

Peer Referrals Do Not Determine the Strength of Population Inference

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.