cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Choose Language Hide Translation Bar

Tolerance Intervals for Mixed Effects Models

What inspired this wish list request? 
I added this to the wish list because while JMP already supports tolerance intervals for simple models and prediction intervals for mixed effects models, there’s currently no way to compute tolerance intervals for mixed effects models. This limitation is especially relevant in quality and regulatory contexts, where users need to make predictions or statements about the proportion of future outcomes. Prediction limits can be unreliable in small sample sizes.

 

What is the improvement you would like to see? 
I would like JMP to have the option to calculate tolerance intervals (with formula included) directly from the fit model platform for mixed effects models, with the option to specify coverage and confidence. Additionally, they would be shown in the prediction profiler as well.

Why is this idea important? 

Tolerance limits are important because they allow us to make probabilistic statements about the proportion of a population that falls within some specified limits, not just individual observations. They are essential in fields where quality is critical, such as pharma or medical devices. It is very important that these intervals work within the mixed effects models framework, since the a lot of data in these fields involves random effects like batches, operators, or sites, which need to be properly accounted for to ensure accurate and reliable results.

5 Comments
jszarka
Level V

Which "simple models" are tolerance intervals currently supported in for JMP?

My assumption is that implementing the same formulas as what is offered in Design Expert would be acceptable (Design Expert Tolerance Interval Formula) although that may only be for multivariate regression with fixed effects and may need more sophistication for mixed effects models.

jszarka
Level V

I did find a Design Expert example with mixed models where a tolerance interval is calculated in their Point Prediction node (Design Expert - RSM with Random Blocks).

ThomasB
Level II

@jszarka Actually, I only recall seeing tolerance intervals in the distribution platform, so the simple model is just the mean model (y_i = mu + e_i). 
There is a formula for tolerance intervals in mixed models in: Francq BGLin DHoyer WConfidence, prediction, and tolerance in linear mixed modelsStatistics in Medicine20193856035622https://doi.org/10.1002/sim.8386. I believe JMPs prediction intervals for mixed effects are also based off of this reference. The implementation is definitely complicated, though.

SarahGilyard
Staff
Status changed to: Under Consideration

Thanks for posting Thomas! As we discussed, this capability is currently under research for possible inclusion in our future roadmap. 

SarahGilyard
Staff

Correct, ThomasB. Currently in JMP we have Tolerance intervals in the distribution platform for normal distribution and non-parametric. In JMP 19, we have added additional distributions such as Log Normal, Gamma, Exponential, and Weibull. Your reference paper is also correct for the source of the prediction intervals in linear mixed models.  Thanks!