The Use of Composite Estimators with Two Stage Repeated Sample Designs

D. Holt and T. Farver

Abstract:
The use of Composite Estimation for inference from repeated survey designs has a long history. Whilst algebraic extensions have been made to take account of two stage sampling designs, there does not appear to have been a systematic attempt to understand how the clustered structure of the population affects the potential gains from composite estimation. What are the population structure characteristics, for example, that would imply greater or less potential benefits from composite estimation than in the simple random sampling case? A simple superpopulation model is used to characterise a clustered population structure over time and this is used to explore the effect of the population structure on the potential gains in efficiency for composite estimation.

Keywords:
Composite estimation; repeated surveys; two stage design; random effects; superpopulation model.