Fast and robust estimators of variance components in the nested error model Articles uri icon

publication date

  • March 2017

start page

  • 1655

end page

  • 1675

issue

  • 6

volume

  • 27

international standard serial number (ISSN)

  • 0960-3174

electronic international standard serial number (EISSN)

  • 1573-1375

abstract

  • Usual fitting methods for the nested error linearregression model are known to be very sensitive to theeffect of even a single outlier. Robust approaches for theunbalanced nested error model with proved robustness andefficiency properties, such as M-estimators, are typicallyobtained through iterative algorithms. These algorithms areoften computationally intensive and require robust estimatesof the same parameters to start the algorithms, but so far norobust starting values have been proposed for this model. Thispaper proposes computationally fast robust estimators forthe variance components under an unbalanced nested errormodel, based on a simple robustification of the fitting-ofconstantsmethod or Henderson method III. These estimatorscan be used as starting values for other iterative methods. Oursimulations show that they are highly robust to various typesof contamination of different magnitude

keywords

  • Clustered data
    Linear mixed model
    Random effects
    Robust fitting
    Variance estimation