COMPARISON OF EFFICIENCY OF JACKKNIFE AND VARIANCE COMPONENT ESTIMATORS OF STANDARD ERRORS

ABSTRACT Large scale surveys usually employ a complex sampling design, and as a consequence no standard methods for estimation of the standard errors associated with the estimates of population means are available. Resampling methods, such as jackknife or bootstrap are often used, with reference to their properties of robustness and reduction of bias. We examine a method based on variance component models as an alternative to the jackknife procedure used for calculation of the standard errors for the subpopulation means of proficiency scores in a large scale survey of education in the U.S.A.