I have a pharmaceutical company client who is asking a tolerance interval question. I know what tolerance intervals are, but this is beyond my capabilities. Hoping someone out there can help and save me some research time.
The client's question started as this:
I am interested in using R to calculate tolerance intervals for a percentile for normal distributions and possibly non-parametric intervals. These are covered in the new version of Meeker and Hahn’s Statistical Intervals text. Can you recommend a particular implementation that works best with JMP?
To which I replied:
- JMP does compute tolerance intervals, and includes the non-parametric calculations by Meeker & Hahn (2017). See info in the Help section of JMP here and here. Perhaps that’s the same as the R calculations?
- If you are interested in connecting JMP directly to R, here is a nice video describing the process. You can pass data and results between JMP and R, and call specific R routines. Note that apparently R needs to be installed on your local machine for this communication to work.
- Here is the most-recent set of JMP/R Integration commands (available in JMP Help)
- If you are asking about which version of R works best with JMP, I don’t know of any limitations. I happen to be using “TKIntRJMP.R version 14.0“ on my Windows 10 machine, and haven’t had any problems.
To which the client replied:
Hahn and Meeker reference R functions which can compute tolerance intervals for order statistics (e.g. 10th percentile). Ultimately, I am trying to understand the mathematical basis of the attached ASTM standards. They apply a tolerance interval to a computed sample statistic (Acceptance Value) under the joint distribution of sample mean and standard deviation for various sample sizes and varying confidence levels and coverage. See attached papers. If this could be implemented in JMP that would be huge. Currently this has been made available via a SAS program (https://ispe.org/initiatives/blend-uniformity-content-uniformity/tools) written by the authors.
(I'm attaching the referenced papers to this post.)
So I've got several questions for the group, and advice on any of them would be welcome:
- Does anyone have experience with these methods? (If so, I can take further discussion off line with you.)
- Is this even appropriate for an SE to be digging into, or should it go elsewhere? (And where would that be?)
- Thoughts on whether this would be something good to suggest for future versions? Would other clients be interested?
Thanks!
Jerry
