Abstract:
|
Clustered survey data are commonly obtained from a multi-stage sampling design, so many methods propose the use of random effects to represent cluster-specific characteristics in the model. Because they often assume the independence among random effects, they cannot represent inter-cluster correlation. However, survey cluster with spatial information is important; for example, spatial correlation in small area sampling. Ignoring correlation structure could give inefficient analysis. In this talk, we introduce the hierarchical (h-) likelihood method for the survey data analysis to accommodate correlated random effects. We also investigate the weighted h-likelihood method to consider the informative sampling design. Small simulation study is carried out to investigate the performances of the proposed methods. Real data examples illustrate the practical use in survey sampling.
|