Seminar Bias control for unit-level M-quantile-based small area estimators

Speaker: Gaia Bertarelli (Università Cà Foscari)

Abstract:

Projective outlier-robust M-quantile-based small area estimators can be substantially biased when the sample data contain representative outliers. Two new predictive-type bias-corrected versions of these estimators are presented for continuous and discrete outcomes. In the presence of both area-level and individual-level outliers in the population, these estimators are more efficient than the robust-predictive and robust-projective estimators proposed in the small area estimation literature. Two estimators of the prediction mean squared error are also introduced: one based on Taylor linearization and the other based on a semi-parametric bootstrap method. The theoretical properties of the proposed methods are illustrated through empirical evidence and supported by simulation studies. An application to the estimation of average income and unemployment rates for local labour market areas in Italy further shows the practical relevance of the approach.

Organizer: Aldo Gardini