Snowball sampling is a non-probability technique in which initial participants (“seeds”) refer the researcher to other potential participants from within their networks, who then refer additional participants, creating a “snowball” effect. It is widely used when no reliable sampling frame exists for the target population.
Snowball sampling fits exploratory and qualitative study designs, particularly when studying sensitive topics or populations without an accessible list (sampling frame). It also appears in mixed-methods designs and, in its respondent-driven form, in quantitative designs aiming for some degree of statistical inference.
Purpose: When and Why to Use It
- Used when the target population is hidden, stigmatized, or otherwise difficult to identify through standard sampling frames.
- Appropriate for studying close-knit or networked communities (e.g., professional networks, support groups).
- Useful in early, exploratory phases of a research question where population boundaries are unclear.
- Helpful when trust and referral from existing participants is needed to gain access.
Fit with Quantitative, Qualitative, and Mixed-Methods Research
| Approach | Typical Role of Snowball Sampling | Example |
| Qualitative | Common method for accessing hidden populations for interviews or focus groups | Interviewing undocumented workers via trusted referrals |
| Quantitative | Used cautiously, often via respondent-driven sampling with statistical adjustments | Estimating prevalence of a behavior in a hidden population using RDS weighting |
| Mixed Methods | Recruits an initial qualitative sample that can inform a later, more structured quantitative phase | Interviewing a small referral sample, then designing a survey based on emergent themes |
How It Works
Step-by-Step Process
- Define the research question and the hard-to-reach population of interest.
- Identify and recruit one or more well-connected “seed” participants.
- Collect data from seeds and ask them to refer other eligible participants.
- Recruit and collect data from referred participants (“wave 2”).
- Continue through successive referral waves until saturation or target sample size is reached.
- Track the referral chain to assess diversity and potential bias in the resulting sample.
Types and Variations
| Type | Description |
| Linear Snowball | Each participant refers only one additional participant, forming a single chain. |
| Exponential Discriminative | Each participant refers multiple contacts, but only some are selected based on criteria. |
| Exponential Non-Discriminative | Each participant refers multiple contacts, and all are included in the sample. |
| Respondent-Driven Sampling (RDS) | A formalized variant using referral coupons and statistical weighting to approximate probability sampling. |
Strengths and Limitations
Strengths
- Provides access to populations that are otherwise very difficult or impossible to sample.
- Builds trust through peer referral, which can improve participation rates.
- Cost-effective and requires no formal sampling frame.
- Flexible and adaptable as the researcher learns more about the population.
Limitations
- Prone to homophily bias, since referred participants often resemble the original seeds.
- Sampling frame is unknown, making it difficult to calculate selection probabilities.
- Final sample size is hard to predict or control in advance.
- Raises privacy and confidentiality considerations, since participants are naming others.
Effect on Internal and External Validity
| Validity Type | Typical Impact |
| Internal Validity | Can be reasonably strong for studying the specific network or population reached, since participants are directly relevant to the research question. |
| External Validity | Typically weak, because referral networks are not representative of the broader population and homophily skews who gets included. |
Sample Size, Effect Size, and Study Power
Because the sampling frame is unknown in advance, traditional probability-based sample size and study power formulas are difficult to apply directly to standard snowball sampling.
- Qualitative studies typically stop recruitment once data saturation is reached across referral waves.
- Researchers often set a target number of referral waves (e.g., three to four) rather than a fixed sample size.
- Respondent-driven sampling (RDS) allows researchers to apply statistical weighting, enabling rough estimates of effect size and supporting limited inferential analysis.
- Whatever the approach, researchers should report how saturation, network diversity, or weighting was assessed.
Guidance by Academic Level
For Undergraduate Students
- Use snowball sampling for small qualitative projects involving niche or close-knit communities, such as a student club or local interest group.
- Keep referral chains short (two to three waves) given typical project timelines.
- Clearly explain in your methods section why a hidden or hard-to-reach population required this approach.
- Pay close attention to informed consent and confidentiality, since participants may be naming peers.
For Graduate Students
- Consider respondent-driven sampling if your research question requires population-level estimates, not just description.
- Track and report referral chain length, diversity, and homophily as part of your methodological rigor.
- Discuss the trade-off between access (a major strength) and external validity (a major limitation) explicitly in your thesis or dissertation.
- Coordinate with your IRB or ethics board early, since chain-referral recruitment of sensitive populations often requires extra protections.
Implementation Checklist
- Define the hidden or hard-to-reach population and the research question.
- Recruit one or more diverse seed participants.
- Develop a referral protocol and consent process.
- Track each referral wave and participant characteristics.
- Monitor for saturation or diminishing diversity across waves.
- Report referral chain structure and limitations in the write-up.
Common Mistakes to Avoid
- Relying on a single, narrow seed group, which amplifies homophily bias.
- Failing to track referral waves, making it impossible to assess sample diversity later.
- Overlooking confidentiality risks when participants name other individuals.
- Presenting findings as broadly generalizable despite the non-random referral process.
Frequently Asked Questions
When should I use snowball sampling instead of another non-probability method?
Choose snowball sampling specifically when no sampling frame exists and the population is hidden, stigmatized, or otherwise hard to access through direct recruitment. If the population is easily accessible, convenience or purposive sampling may be more efficient.
Does snowball sampling always produce biased results?
It carries a structural risk of bias due to homophily in referral networks, but the severity depends on network diversity and the number of independent seeds used. Using multiple, diverse seeds and tracking referral waves can reduce, though not eliminate, this bias.
Can I use snowball sampling for a quantitative study?
Yes, particularly through respondent-driven sampling (RDS), which applies statistical weighting to referral data to support more defensible quantitative inference, though it remains more complex than standard probability sampling.
How do I decide when to stop recruiting?
Most researchers stop when data saturation is reached, when referral waves stop introducing new perspectives, or when a pre-set target number of waves or participants has been achieved.
This article was originally published on January 30, 2025, and updated on June 14, 2026.
