|
Abstract:
|
Sampling populations with rare, hidden, or hard-to-reach traits often requires innovative methods such as link-tracing techniques, in which social links are followed to bring more members of the hidden population or more persons with the rare traits into the sample. Although such methods can be effective, they require inference methods that, in some respects, compensate for the design to give an unbiased or accurate picture of the population. Efficient design-based approaches involve Rao-Blackwell or Markov chain resampling methods. The most effective model-based approaches involve likelihood or Bayes methods and generally involve Markov chain Monte Carlo methods for feasible computation. A third approach is to use a targeted random walk design to obtain simple estimates that are, in a sense, robust against the lack of control of some aspects.
|
