Introduction

It is clear that surveying difficult-to-reach populations is a universal problem (Maranda, 2004; Statistics Canada, 2004).

It is universal in several senses. First, some populations, subgroups, and individuals will always be hard to survey. Second, there are diverse and complex reasons that populations and individuals are challenging. Difficulty is neither simple nor unidimensional. Finally, difficulty spans the globe. Wherever surveys are done, difficulties arise. But the specific mix of impediments can be society-specific and the steps needed to overcome these impediments must be geared to the realities and complications that prevail in each survey in every country.

The total survey error paradigm can be used to examine the challenges in surveying hard-to-survey populations. Major components of total survey error that specifically relate to these populations include (1) sampling problems, such as the need for special sample frames, (2) noncoverage or undercoverage, (3) nonresponse, and (4) measurement errors, such as misreporting.

First, sampling problems include subgroups that are part of the target population, but that are not covered or undercovered by available sample frames. If the subgroup is the focus of the study, an alternative sampling design often must be developed. Even if the target population is not undercovered, using a sampling frame for the general population and screening down to the targeted subgroup may be inefficient and impractical due to cost. Dual or multiple frames or various referral sampling methods will often be needed to sample adequately and efficiently rare populations (Christman, 2009; Elliott, McCaffrey, Perlman, Marshall, & Hambarsoomians, 2009; Ericksen, 1976; Johnston & Sabin, 2010; Kalsbeek, 2003; Kalton, Chapter 19 in this volume; Kalton & Anderson, 1986; Reed, 1975–76; Rothbart, Fine, & Sudman, 1982; Sudman, 1972). What can be done for a given target population will vary greatly from country to country depending on what information is available for developing sampling frames (McKenzie & Mistianen, 2009; Salganik & Heckathorn, 2004; Treiman, Lu, & Qi, 2009). For example, if a national census or administrative records fail to identity a particular subgroup or rare population, then sample frames based on those sources cannot target these groups.