Log-normal process capability

[#1]

Hi, I have a process that is log-normally distributed and has an upper spec limit, and I'd like to calculate a Cpk. I tried two approaches that gave me different answers:

1. Save spec limit as column property for process. Plot distribution of process, then Continuous Fit --> LogNormal to get a Cpk (checking goodness-of-fit of course)
2. Add new column. Set column formula as log(process), save column upper spec limit as log(upper spec limit). Plot distribution, get Cpk from capability analysis. Validate via Continuous Fit --> Normal and checking goodness-of-fit.

Which of these approaches would you recommend? Or could you suggest a better approach?

Many thanks,

Alex

[#2 · Accepted solution]

I consider both of the techniques you specified as good vialble processes.  However, the process that I would actually recommend is a very nice builtin capability of the Distribution Platform.  If you run the Distribution Platform, you can then go to

     Continuous Fit==>All

It will then evaluate your data to find which distribution(s) it fits.  Once done, and if a Log Normal distribution is found to be the fit, you can go under "GLot" distribution and

     Save Transform

It creates a new column using the determined Generalized Log algorithm

[#3]

Thanks Jim. It's a bit frustrating that the LogNormal transform can't be saved in the same way - the GLog transform adds additional complexity, which is hard to justify and would be (even) harder to explain to my colleagues.

Any idea why I got different answers between my initial two approaches?

[#4 · Accepted solution]

I understand your frustration.  The question on why log normal does not have a Save Transform would have to be answered by the JMP developer.

Concerning the difference between the Cp/Cpk for the Distribution lognormal and your own transformed to log(x) and then calculate.......it appears that there are differences in the calculation of the PDF.  I am taking that from the description within

     Help==>Books==>Basic Analysis

But a more complete answer should be available from JMP Support

[#5]

I have also been trying to figure out why the Cpk result from using the built-in Capability Analysis and the manual method of calculating the Cpk via a log transform of the data is giving such different results. Any insight into the formula that JMP is using for the log Normal Cpk calculation would be greatly appreciated.