Book Description

Measuring the source and magnitude of components of variation has important applications in industrial, environmental and biological studies. This thesis considers the problem of constructing confidence intervals for variance components in Gaussian mixed linear models. A number of methods based on the usual ANOVA mean squares have been proposed for constructing confidence intervals for variance components in balanced mixed models. Some authors have suggested extending balanced model procedures to unbalanced models by replacing the ANOVA mean squares with mean squares from an unweighted means ANOVA. However, the unweighted means ANOVA is only defined for a few specific mixed models. In Chapter 2 we define a generalization of the unweighted means ANOVA for the three variance component mixed linear model and illustrate how the mean squares from this ANOVA may be used to construct confidence intervals for variance components. Computer simulations indicate that the proposed procedure gives intervals that are generally consistent with the stated confidence level, except in the case of extremely unbalanced designs. A set of statistics that can be used as an alternative to the generalized unweighted mean squares is developed in Chapter 3. The intervals constructed with these statistics have better coverage probability and are often narrower than the intervals constructed with the generalized unweighted mean squares.