Abstract
The exact confidence interval for a ratio of variances can have coverage probability that is much smaller than the nominal level when sampling from nonnormal bivariate distributions. An approximate confidence interval is proposed that performs as well as the exact method in bivariate normal distributions but performs better in bivariate nonnormal distributions. An approximate sample size planning formula is proposed.
Acknowledgements
This research was supported by grant SES-0343552 from the Methodology, Measurement, and Statistics program of the National Science Foundation.