Sparse designs for estimating variance components of nested random factors

Abstract

A new class of designs is introduced for both estimating the variance components of nested factors and testing hypotheses about those variance components. These designs are flexible, and can be chosen so that the degrees of freedom are more evenly spread among the factors than they are in balanced nested designs. The variances of the estimators are smaller than those in stair nested designs of comparable size. The mean squares used in the estimation process are mutually independent, which avoids some of the problems with staggered nested designs.