Estimation of the smallest normal variance with applications to variance components models

It is proved that improved point and interval estimators for the smallest normal variance can be directly obtained from the unrestricted normal variance estimation counterparts. Applications to estimating the error variance in general random or mixed effects models are given.