Online Program

Sampling elusive and mobile populations

*Graham Kalton, Westat 

Keywords: location sampling, respondent-driven sampling, snowball sampling

Achieving efficient probability samples of elusive and mobile populations presents a major challenge in survey research. Elusive populations, such as men who have sex with men, injection drug users, and illegal immigrants, are difficult to sample both because of their rarity in the general population and often the sensitivity of identification. Mobile populations such as the homeless and nomads are difficult to sample because they are not covered in regular household surveys. Visitors to certain locations (e.g., national parks and museums) and travelers (e.g., persons entering a country and car passengers)are also mobile populations that are generally best sampled and surveyed on location. This paper will examine the application and limitations of standard probability sampling methods for sampling such populations,including large scale screening, disproportionate stratification,two-phase sample, multiple frames,and network sampling. It will then review such methods as location sampling, snowball sampling, and respondent-driven sampling for sampling elusive and mobile populations, methods often used for such populations in both developed and developing countries. Location sampling, also known as venue-based or time-space sampling, covers only those who visit at least one of the specified locations in the survey time period. Weights are needed with location sampling to compensate for unequal selection probabilities of visits and, for inferences about the visitor population, for the number of visits a sampled person makes during that period (in practice often not applied). Estimating a sampled person's number of visits is not straightforward. Variability in the standard weights can cause a serious loss of precision, resulting in the need for some procedure for reducing the variability in the weights. Inferences from snowball and respondent-driven samples are based on a number of assumptions that need to be critically assessed.