Abstract:
|
Informative sampling is common in survey sampling and this leads to selection effects. Building on the extensive frequentist literature, Bayesian methods may provide added benefits. We present an organizational structure that should help to identify areas where additional research is needed. Three factors are considered: (a)the finite population indicator function is known vs. only the indices of the sampled units are known, (b)the selection process is based on a parameterized superpopulation model or on a covariate, and (c)inference is for finite population quantities or superpopulation parameters. Finally, several specific findings, at variance with current methodology, will be described.
|