03 · What You Need to Know
What Makes Snowball Sampling Methodologically Defensible?
What Exactly Is Snowball Sampling?
Snowball sampling is a recruitment approach in which initially identified participants or contacts help researchers locate additional potentially eligible participants. Those additional participants may subsequently identify further candidates, allowing recruitment to expand through referral chains.
It is also commonly described as chain-referral sampling, although terminology and implementation vary across methodological traditions.
For example, a researcher investigating informal professional learning networks among teachers may begin with several educators known to participate in such networks. Those educators may identify colleagues who have relevant experiences but are not visible through institutional directories.
Unlike simple random sampling, ordinary snowball sampling does not give every member of the target population a known probability of selection.
Recruitment depends partly on social relationships, participants' willingness to refer others, and the accessibility of particular networks.
These features can be useful for identifying participants, but they also shape the resulting sample.
When Is Snowball Sampling Particularly Appropriate?
Snowball sampling may be appropriate when a population is difficult to enumerate or when membership is not readily identifiable through public or administrative records.
This can occur when participants share specialized experiences, belong to informal professional networks, or are reluctant to disclose their involvement in a particular activity through conventional recruitment channels.
For example, researchers may use referrals to identify educators who have participated in informal communities of practice, professionals involved in specialized institutional processes, or individuals with uncommon experiences relevant to a qualitative investigation.
Snowball sampling may also help establish initial contact when trust and professional relationships influence willingness to participate.
However, difficulty recruiting participants does not automatically justify referral sampling. Researchers should consider whether institutional records, open recruitment, purposive sampling, or other methods could identify the required cases more directly.
Referral recruitment is most defensible when social connections provide a meaningful access route that other feasible methods cannot readily supply.
How Does Snowball Sampling Actually Work?
A typical procedure begins with a small number of initial participants, sometimes called seeds. These individuals satisfy the study's eligibility criteria and provide potential routes to additional participants.
The researcher then invites referrals to participate, verifies their eligibility, and may ask them to identify further candidates.
Step 1: Define the target participants.
Establish the experiences, characteristics, or circumstances required for inclusion.
Step 2: Identify initial participants.
Recruit suitable seeds through defensible channels, preferably across different relevant networks where feasible.
Step 3: Establish a safe referral procedure.
Explain how participants may share study information or identify potential candidates without compromising privacy.
Step 4: Screen referred candidates.
Verify that interested individuals meet the eligibility criteria and obtain their informed consent independently.
Step 5: Continue recruitment selectively.
Evaluate whether additional referral chains contribute relevant perspectives or merely repeat the same network.
Step 6: Document the recruitment process.
Record referral waves, recruitment channels, relevant network limitations, and the rationale for stopping recruitment.
The process should not operate as an uncontrolled expansion of contacts. Researchers remain responsible for determining eligibility, managing recruitment, and ensuring that participation is voluntary.
How Should Initial Participants Be Selected?
The initial participants can strongly influence the eventual composition of a snowball sample.
If all seeds belong to the same institution, professional association, or friendship network, subsequent referrals may remain concentrated within that network.
For this reason, researchers may deliberately recruit initial participants from different relevant settings or backgrounds.
For example, a study of informal AI teaching communities might begin with educators from public and private institutions, different disciplines, and different geographical locations.
This approach can broaden access to potentially different referral networks. However, it does not guarantee that all relevant networks or population groups will be reached.
The rationale for selecting initial participants should be explicit, particularly when their professional standing or institutional affiliations may influence subsequent referrals.
What Is Network Homophily, and Why Does It Matter?
Homophily refers to the tendency for social connections to occur more frequently among people who share particular characteristics, experiences, or circumstances.
In referral sampling, this means that participants may recommend people similar to themselves.
For example, teachers who actively support generative AI may refer colleagues who share their enthusiasm. Teachers who oppose its use may belong to different professional networks and remain unrepresented.
Consequently, the final sample may overrepresent particular attitudes, institutions, or professional communities.
This problem is not necessarily visible from the number of participants recruited. A sample of 50 people drawn through one closely connected network may contain less diversity of relevant experiences than a smaller sample recruited across several distinct networks.
Researchers should therefore consider how network structure affects who can be identified and invited.
Does Using Multiple Referral Chains Reduce Network Bias?
Beginning with participants from different networks may reduce dependence on a single recruitment pathway and broaden the range of experiences encountered.
For example, a researcher might establish separate initial referral chains among educators from different institutional sectors or geographical areas.
However, several chains may still overlap or lead toward the same highly connected participants.
Researchers can monitor recruitment patterns and assess whether new referrals are introducing substantively different cases.
Where important perspectives remain absent, additional purposive recruitment may be justified.
These measures can improve the breadth of the observed sample, but they do not establish known population inclusion probabilities.
Network diversity should therefore be treated as a design consideration rather than proof of statistical representativeness.
How Is Snowball Sampling Different from Purposive Sampling?
Purposive sampling and snowball sampling can operate together, but they describe different aspects of recruitment.
Purposive sampling
Specifies why particular participants or cases are relevant to the research question.
Snowball sampling
Specifies how additional potential participants are identified through referrals and social connections.
A researcher may establish purposive eligibility criteria and use snowball referrals to locate people who satisfy them.
For example, only teachers who have participated in informal AI-related professional learning communities may be eligible. Existing participants then help identify other eligible teachers.
In that arrangement, purposive criteria determine relevance while snowball recruitment provides access.
Neither method should be described as probability sampling merely because recruitment follows a systematic procedure.
Is Snowball Sampling the Same as Respondent-Driven Sampling?
No. Although both involve peer referrals, respondent-driven sampling, or RDS, is a specialized network-based method with additional recruitment and estimation procedures.
RDS commonly uses structured recruitment chains, controlled referral opportunities, and information about participants' network connections. Some RDS estimators attempt population inference under assumptions about network structure, recruitment, and sampling behavior.
Ordinary snowball sampling generally does not collect or use the information required for those estimators.
Even RDS estimates can be sensitive to violations of their assumptions, including network connectivity, recruitment behavior, and seed dependence.
Researchers should therefore avoid describing ordinary referral recruitment as respondent-driven sampling simply because participants recruit other participants.
Nor should they assume that extending referral chains automatically makes the sample representative.
Can Snowball Sampling Estimate Population Prevalence?
Ordinary snowball sampling generally cannot support conventional design-based prevalence estimation because population inclusion probabilities are unknown.
For example, if 70% of recruited educators report using generative AI, that percentage describes the observed respondents. It does not establish that 70% of all eligible educators use generative AI.
People with larger professional networks, stronger interest in the subject, or greater willingness to refer others may have more opportunities to enter the sample.
Those differences can affect the observed distribution of outcomes.
Population estimation from non-probability samples may sometimes be attempted using additional models, auxiliary information, or specialized sampling designs, but the assumptions require explicit justification.
The limitations of estimating population prevalence from non-probability samples are especially relevant when interpreting referral-based data.
What Ethical Issues Are Specific to Snowball Sampling?
Referral recruitment can create privacy risks because one participant may disclose another person's identity, contact information, or involvement in a sensitive activity.
Researchers should not assume that consent to participate includes permission to identify colleagues, friends, or other contacts.
A relatively privacy-protective approach is to provide participants with study information that they may voluntarily forward to potentially eligible individuals. Interested recipients can then contact the research team directly.
Where direct referrals are necessary, the procedure should be consistent with ethics approval, applicable privacy requirements, and an appropriate basis for processing contact information.
Researchers should also consider whether participants may feel pressured to participate because a supervisor, colleague, or influential network member recommended them.
Watch Out
Do not disclose a participant's identity or participation status to other members of a referral network without appropriate authorization. In sensitive studies, even revealing that someone may be eligible can expose private information.
Referral procedures should protect both existing participants and people who have not yet agreed to participate.
How Many Referral Waves Are Necessary?
There is no universal number of referral waves required for defensible snowball sampling.
A referral wave refers to a stage of recruitment in which participants identify potential candidates who may subsequently refer others.
Additional waves can extend recruitment beyond the original seeds, but they do not necessarily produce broader population coverage.
Some chains may terminate quickly, while others may continue through highly connected communities.
The appropriate stopping point depends on the research objective, analytical adequacy, ethical constraints, and whether additional recruitment is likely to contribute relevant evidence.
For qualitative studies, the researcher should evaluate sample adequacy in relation to the analytical approach rather than assuming that a fixed number of waves or interviews establishes saturation.
How Can Researchers Strengthen a Snowball Sampling Design?
Several design practices can reduce avoidable weaknesses without eliminating the method's non-probability character.
Researchers can begin with diverse initial participants, establish consistent eligibility criteria, monitor referral chains, and supplement referrals with other defensible recruitment channels when important perspectives remain inaccessible.
It is also useful to record whether referrals repeatedly originate from the same institution or professional network.
When the research question requires contrasting perspectives, researchers should deliberately examine whether recruitment is dominated by participants with similar experiences or positions.
These practices are consistent with broader efforts to reduce the limitations of non-probability sampling.
Nevertheless, improved recruitment diversity does not establish statistical representativeness or eliminate unobserved network selection effects.
How Should Snowball Sampling Be Reported?
The methodology should explain why referral recruitment was necessary, how initial participants were identified, how referrals were requested, and how eligibility was verified.
Researchers should also describe whether multiple recruitment chains were used, whether referrals crossed institutional or social boundaries, and how recruitment was stopped.
Where confidentiality permits, reporting the number of seeds, referral waves, and participants recruited through different pathways can help readers understand the sample's composition.
For qualitative interview studies, reporting frameworks such as COREQ and the Standards for Reporting Qualitative Research can support transparent descriptions of recruitment and participant selection.
The explanation should distinguish the value of referral-based access from the limits of population inference.
