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