Respondent-driven sampling is a form of snowball sampling which allows researchers to make unbiased estimates about hidden population like injection drug users and sex workers. It can be used to answer questions like what percentage of drug injectors in New York City share needles? Because of the many advantages of respondent-driven sampling, it is currently be used by the U.S. Center for Disease Control and Prevention (CDC) and other international health organizations.

A respondent-driven sample is collected via a snowballing design, where current subjects recruit their friends to be future subjects. Because subjects are not selected via simple random sampling, care must be taken when making estimates from this type of sample. However, it was recently shown that if the appropriate estimation procedures are used and if certain conditions are met, then prevalence estimates from this type of sample are unbiased (Salganik and Heckathorn, 2004).

One of the main 'tricks' to getting unbiased estimates from a snowball sample is to re-think the normal sampling and estimation process. Ordinarily, we use a sample to directly make inference about a population. However, as is well documented in the literature, snowball samples are very badly suited for estimation within this framework. On the other hand, snowball samples are very well suited to make estimates about social networks. In fact, when snowball samples were first introduced by the sociologist James Coleman, they were intended for making estimates about social networks and not populations. Taking advantage of the special nature of the snowball sample, the respondent-driven sampling estimation procedure uses the sample to make estimates about the social network connecting the hidden population. Then the characteristics of these networks are used to estimate the prevalence of a specific trait. This process can be seen in the schematic below.