Small area estimation using both survey and census unit record data: Links, alternatives, and the central roles of regression and contextual variables

This chapter focuses on the central role played in small area estimation (SAE) by choice of contextual variables to control area-specific bias and regression variables to reduce standard errors. The marked reduction in standard errors in models such as ELL which use both unit record survey and census data, when they are compared with some other SAE methods, reflects in part an apparent focus in the published statistical literature on predicting rather than minimising random effects and standard errors. The chapter also highlights the central importance of sound and careful choice of regressors and contextual variables to minimise standard error and to control bias in any good model, whatever the choice of SAE method. Spatial models are an accepted part of mainstream statistical methods. Accuracy of small area estimates would be improved if sample surveys used for modelling were designed with small area estimates in mind.