75776_en-US

liliconcom

Sample size, power and TOST for a comparability/equivalence test against a gold standard

I have a historical dataset as a baseline (key product measure) for a manufacturing process, then the process has changed due to optimization. How do I: 1)  estimate sample size and power for a comparability study? 2) use TOST to establish if the product from the optimized process produced the same/comparable product? (assuming equal variance between the 2 datasets)

 

For the 1) my current approach is to use std dev from the baseline dataset, set alfa = 0.05 and 'difference in mean' using delta between 'baseline mean' and 'the failure value/lower spec' , from that I can chose a power to get the sample size from DOE>Design and Diagnostics>Sample Size and Power>Two Sample Means. Is this approach correct?

 

For the 2) I have never done it so I am seeking an expert answer please.   

 

Thanks a lot,

Ed

Re: Sample size, power and TOST for a comparability/equivalence test against a gold standardThanksRe: Sample size, power and TOST for a comparability/equivalence test against a gold standard

I can't tell if you are confirming my previous reply or asking for clarification. If you were asking for more help with TOST in JMP, see the reference I gave earlier for Oneway. It is all there with examples. Please post any further questions that you might have.

Re: Sample size, power and TOST for a comparability/equivalence test against a gold standardRight. what is the right approach to obtain tolerance interval in my case and how to use it in TOST?. I have just one failure spec limit and a historical dataset that reflects key product potency and releases the product from the old process.

Thanks again for your reply.
EdRe: Sample size, power and TOST for a comparability/equivalence test against a gold standard

That is correct. The TOST is for the population mean. So is the (Target - Spec) interval reasonable for the tolerance in the test?

Re: Sample size, power and TOST for a comparability/equivalence test against a gold standardThanks for the reply. I think I am still looking for a population comparison with TOST, given that I can get a dataset with valid sample size and study power.Re: Sample size, power and TOST for a comparability/equivalence test against a gold standard

See Help > Books > Design of Experiments Guide > Prospective Sample Size and Power chapter. Your definition of the similarity interval is not unreasonable. However, if the mean has shifted to the specification, you might determine that the equivalence is significant while experiencing many individual units out of specification. The specification is about each unit. The mean is about the population. The equivalence test is about the population.

 

See Help > Books > Basic Analysis > Oneway Analysis chapter.