804102_en-US

liliconcom

Tolerance intervals for linear regression

Howdy SEs:

Have you ever had a customer request for tolerance intervals on regression predictions? One of my customers is very persistent on this topic, and he insists that the prediction intervals we recently added to the profiler are not sufficient for his needs. 

I'm referring him to the Wish List, just curious if anybody else has seen this request.

By the way, here's an interesting Stack Exchange thread about the difference between prediction intervals and tolerance intervals: https://stats.stackexchange.com/questions/26702/prediction-and-tolerance-intervals

Re: Tolerance intervals for linear regression

Bill Kappele, link, teaches the use of tolerance intervals as well as prediction intervals as part of his DOE curriculum. JMP can calculate them in the distribution platform, but they aren't in the profiler, as you know. I've used them a fair bit back when I was in the real world.

Re: Tolerance intervals for linear regression

In my Amgen days, I recall having a discussion on whether we should be using confidence intervals or tolerance intervals for establishing/confirming shelf-life. Ultimately, we decided that CIs made more sense since we were interested in the trend and not the individual observations making up the trend. That said, I could image others would feel the opposite and would want the tolerance intervals so they could make a statement about a given shelf-life covering a proportion of the population.