ANOVA-type estimators of variance components for nested error regression models are always constructed based on moment equations related to residual variance. We consider moment equations associated with covariance and construct improved ANOVA-type estimators. These estimators are seen to be consistent, asymptotically unbiased and have better performances than traditional estimators of variance components for almost all kinds of sample allocations. Their improved performance is demonstrated analytically as well as through detailed simulation studies and applications to real data sets.
Joint work with Tatsuya Kubokawa, Faculty of Economics, University of Tokyo and Shonosuke Sugasawa, Center for Spatial Information Science (CSIS), The University of Tokyo, Japan.