Abstract

Many developed countries have high quality census data and/or population registers that can be used to build sampling frames for surveys. In other countries, however, census data is out of date or traditional sampling methods are impractical or dangerous. Multinational surveys very often include one or more countries where traditional sample designs do not work. Problems may occur at the first design stage, in which clusters are selected with probability proportional to size, due to out of date or unavailable census data. Problems can also arise at later design stages, such as persons or households selected within the clusters, because no register data are available or listing households within the cluster is not feasible. This chapter describes the options that are available to samplers in such situations. Techniques to be discussed include: random geographic cluster sampling and nighttime lights at the first stage; and reverse geocoding, random walk, respondent-driven sampling, and quota sampling at a subsequent stage. For each method, we describe the statistical properties and note the pros and cons. Throughout, we suggest the best sampling techniques as ones that minimize interviewer discretion and contain built-in opportunities for verification of interviewer performance.