Hi,
I got an example from a customer. He uses Process Capability Analysis and compares it to Minitab. He gets different results for within sigma (higher CPK in JMP), which makes it impossible to use JMP as there is the risk that they trust a process more than they should based on Automotive guidelines (ISO/TS, IATF).
But this is not the worst problem. When I tried to find out what happens, I got confused by the different behaviours in JMP itself:
* Using PCA plattform I saw that you can set the subgroup size and then change the statistics. As Minitab using a pooled standard deviation with an unbiased estimator C4(d+1) (based on a Gamma Function), I tried to find the equivalent in JMP. I believe it should be the Average unbiased Stdev, however in JMP's documentation I cannot find any Gamma function used in the formulas, so I believe we do not use a "pooled" stdev, and thats where the difference came in.
* More important: I understand the different results using different statistics (Average of ranges/average unbiased stdev/none). What I do not understand is that if I use the Distribution Platform I get other results which are quite close to the MT ones (rounded the same). Whatever option I take in PCA plattform I do not get to similar results. What do I do wrong?
* Side note: IN the distribution platform you can only show within sigma, when there are no spec limits in the column properties. Otherwise you only get the overall sigma and no option to add within sigma afterwards. If there is a trick for that, please let me know as well, otherwise I will add a defect for this.
Attached the Powerpoint and an example file with scripts reproducing the analysis. The important columns are "Value 1" and "Value 1 - wo spec", which are the same expect for column property spec limits (missing in the latter). Subgroup size is 5, and I know this is not best normal distribution as it has a tail, but the calculation should still be similar, shouldn't it?
Thanks
Martin
