Introduction
While there are difficulties in precisely defining hard-to-survey (H2S), hard-to-find, rare, hidden, or elusive populations (see Chapter 1 for a discussion of definitions), there is general agreement that it is very difficult to locate certain population subgroups. Studying such groups using cross-sectional surveys of the general population is challenging, because sample sizes are often too small to provide reasonable precision for point estimates and statistical power for comparisons. If the H2S groups are the target population of a study, sampling of their members becomes an issue.
Sampling for scientific data collection with these H2S populations is one of the most notable challenges discussed in the sampling literature (e.g., Kalton, 2009; Sudman, Sirken, & Cowan, 1988). Some studies mistakenly argue that frames do not exist for these populations (e.g., Paquette & de Wit, 2010). It is true that there are no readily available sampling frames exclusively of these population members. However, it is technically possible to sample from the general population and screen for the target population members. This type of screening presents two challenges. First, building a sampling frame for such H2S populations is costly. Assume that a study has HIV positive cigarette smokers as its target population, as in Humfleet, Delucchi, Kelley, Hall, Dilley, and Harrison (2009). A large number of households sampled from the general population need to be screened to find enough people who meet the criteria of both being HIV positive and smoking cigarettes. Second, depending on the level of social stigma and discrimination associated with the H2S population of interest, some population members may misreport their eligibility intentionally in the screening interviews in order not to reveal their identity. HIV positive cigarette smokers are associated with socially stigmatized HIV status as well as the socially undesirable status of being a smoker. For these reasons, traditional probability sampling approaches, although ideal, are often regarded as being infeasible and impractical. Although expensive to conduct, there are many studies of H2S populations using probability samples (see Table 15.1 of Binson, Blair, Huebner, & Woods, 2007). Examples include the use of multistage area probability samples and random digit dialed (RDD) telephone samples (e.g., Catania, Osmond, Stall, Pollack, Paul, Blower et al., 2001; Cochran & Mays, 2000).
