ABSTRACT
Burdick and Larsen Citation[1] compared the Gui et al. Citation[2] Modified Large-Sample and the Satterthwaite methods for constructing approximate confidence interval for the ratio in a classical gauge R&R study. In this paper, we provide results showing that both methods can fail by a large margin to maintain the nominal confidence level when used to estimate small values of δ. We discuss the implications of these findings and suggest a simple and apparently effective alternative to these methods for estimating δ.
ACKNOWLEDGMENTS
The author thanks Professor Stephen B. Vardeman of Iowa State University for valuable discussion and suggestions. The author was supported by an Oversea Graduate Scholarship from National University of Singapore and a Boeing Dissertation Fellowship from Iowa State University. The simulations were done using the Snedecor Hall Research Computers at the Department of Statistics, Iowa State University.