Comparisons of Random Nested Designs Using Quantile Dispersion Graphs

• Activity Number: 407
• Type: Contributed
• Date/Time: Thursday, August 12, 2004 : 8:30 AM to 10:20 AM
• Sponsor: General Methodology
• Author(s): Juneyoung Lee*+ and Byung-Chul Jung
• Companies: Korea University and Korea University
• Keywords: random model ; nested model ; variance component ; quantiles ; staggered design ; inverted design

Abstract

To compare designs for a linear model, criteria based on single-valued functions are typically used. These functions are, however, often difficult to use not only because of their complex forms but also because of their dependency on the model's unknown variance components. To overcome these limitations, a use of exact quantile values of estimates of variance components have been suggested. The use of quantiles can provide an overall quality of an estimator. When quantiles of estimates of variance components can not be obtained exactly, empirical quantile values can be considered. We focus on a use of quantiles of estimated variance components for a nested random model. To do this, three competitive designs, namely, the two-fold balanced nested design, two-fold staggered nested design, and two-fold inverted nested design are considered. For methods of estimating variance components, ANOVA and ML are used. Graphical approaches known as quantile dispersion graphs and empirical quantile dispersion graphs are employed in order to provide an overall quality of an estimator and to evaluate the quality of estimation obtained with a given design.