Improving confidence intervals on measures of variability in R&R studies