On conditioning for model-based inference in survey sampling
Biometrika, 75, (2), . (doi:10.1093/biomet/75.2.275).
Full text not available from this repository.
A ‘discontinuity’ of sampling distribution inference between a prediction problem and an estimation problem in model-based survey sampling is discussed. The problem is resolved by conditioning on an appropriate ancillary within a curved exponential family framework. The ignorability of the sampling design for inference is discussed.
Actions (login required)