Investigation of the clustering effect in the Belgian Health Interview Survey

Abstract

This paper investigates the effect of clustering in the first Health Interview Survey (HIS) that took place in Belgium in 1997. In this survey 10,221 individuals were interviewed using a stratified multistage clustered sampling procedure. Clustering arises at two levels in the HIS, within municipalities and within households. Its effect and magnitude on some selected continuous and discrete items are studied from a multilevel modeling perspective. This model-based approach fully acknowledges and takes advantage of the hierarchical structure of the data, and is to be contrasted with the more traditional, design-based approach which views the population structure as a nuisance factor. The effect of weighting in this context is also investigated following Pfeffermann et al.