The Double-Edged Sword of Consecutive and Snowball Sampling: Practical Utility Versus Methodological Compromise

The question

What does a research field lose when it recruits subjects by referral, and who ends up missing from the data?

The method

A commentary rather than a study — the authors review the sampling practices of Indian mental-health research and argue from published examples and “lived experiences of over a decade in Indian mental health spaces as a field researcher, clinician, ethics reviewer, and academic supervisor” [1].

The findings

It is genuinely good at what it is for: snowball sampling “can be invaluable for reaching hidden or stigmatized populations, where trust and peer referrals are crucial for recruitment” [2]. It works by referral — it “relies on peer referrals within networks to identify additional participants” [3].

The failure is structural and compounding. “The initial seed participants are often connected by caste, class, religion, or geography. As referrals multiply, the diversity tends to narrow rather than expand” [3]. What comes out the other end looks like a sample and behaves like an echo: “when marginalized groups are represented solely through the most visible or networked members, data reflect only a fraction of lived realities while claiming to be comprehensive” [3]. Their worked example is exact. A study of widows recruited through a self-help group hit its numbers, but “the resulting sample included only those already participating in support networks” [3] — the isolated widows, the ones the study most needed, were the ones the method could not see.

The verdict is not that the method is wrong: “Neither method is inherently unethical, but both demand greater transparency and intentionality than they currently receive” [4].

The limits

This is argument and illustration, not measurement. The examples are drawn from one research literature, and the paper offers no estimate of how much narrowing a referral chain produces. It is a caution with a mechanism, not a quantity.

• You are asking each person on a Calibration Call who else you should talk to. That is snowball sampling, and this paper names its failure mode. Without it, you feel like your list is expanding when it is folding in on itself: person three was nominated by person two, who was nominated by person one, and all of them work the way person one works.
• You are picking your first two or three calls. The paper’s finding puts the weight on the seeds, not the chain. Deliberately unlike seeds — different company stage, different geography, different route into the role — buy you more than any number of additional referrals from one seed.
• You are about to conclude that the calls agreed. Agreement among people from one network is not corroboration. Ask each person how they know the next person before you count their answers as independent.