… standard random sampling is expensive and difficult to achieve in practice even in the best of circumstances.  Snowball sampling has been used in research into marginal or hard-to-reach populations like drug users, gang members, prostitutes, AIDS patients, and other groups.  Often, a referral from a member of the group is the key that unlocks a subject’s willingness to provide information. One early snowball study involved pickpockets; it is hard to imagine randomly selecting pickpockets, much less approaching one out of the blue and saying

“I’m conducting research and would like to ask you about your work as a pickpocket.”

(But see this article about a skilled New York pickpocket who retired from his profession to sell marijuana, at the urging of his wife, who pointed out that it would be safer, legally speaking.)

The difficulty comes when more is claimed for the snowball sample information than the method can support.  It can provide invaluable case-study and exploratory information that could not be gotten in other ways, but it cannot be claimed to generalize reliably – in particular, it is not possible to attach to the results statistical significance or confidence intervals based on random sampling assumptions.