Follow Geordie and don’t do snowball sampling. Don’t do that! Snowball sampling is not random sampling and random sampling is the only procedure that provides optimal possibility for obtaining a representative sample. Unrepresentative samples cannot provide reliable information and should not be used.
Survey researchers know snowball sampling is a nonrandom/nonprobability method of sampling but they continue to use it because they’re addicted to asking and the answers thereby produced. Because they’re addicted, they can’t stop doing what they know they shouldn’t.
Here is a description of snowball sampling sent to me by a university survey research center. I’ve put my comments in the text of this description within brackets and in bold:
“Snowball sampling, also known as chain referral sampling, is a nonprobability [nonrandom–don’t do that!] method of survey sample selection that is commonly used to locate rare or difficult-to-find populations. Although there are several variations, this approach involves a minimum of two stages: (a) the identification of a sample of respondents with characteristic x during the initial stage, and (b) the solicitation of referrals to other potentially eligible respondents [potentially eligible respondents are not random respondents, so… don’t do that!] believed to have characteristic x during subsequent snowball stages. In many applications, this referral process continues (or snowballs) until an acceptable number of eligible respondents have been located. Statistical inferences can be drawn from the first stage of a snowball sample, assuming that probability methods of selection were used. Samples drawn during snowball stages, and samples that combine the initial and snowball stages are not representative, however, and cannot be used to make statistical inferences. [In other words, snowball sampling produces unrepresentative results from which reliable inferences cannot be made, so… don’t do snowball sampling–don’t do that!]
Beyond nonrandom selection procedures, other limitations include correlations between social network size and selection probabilities, reliance on the subjective judgments of informants, and confidentiality concerns [snowball sampling is severely limited and should not be done–don’t do that!].
Key advantages include low cost and the potential time efficiency with which samples can be recruited.” [In other words, it doesn’t cost much to do what you shouldn’t do, so… don’t do snowball sampling–don’t do that!]
For a complete assessment of survey research, see my book, The Problem with Survey Research.