Short Term (Cpk) and Long Term (Ppk) Capability in JMP

[#1]

Hi everyone,

I have a question about the consistency of how JMP displays the Capability and Performance Indices.

I'm going through the STIPS course @mia_stephens  at the moment and, in the Quality Methods Module 3 - Process Capability, it was really clear to me that the software may have inconsistencies in the message with regards to this subject. I have used the Impurity.jmp example to illustrate this. 

 1. If we run a capability analysis through the Distribution plattform for the variable Temp, and leave the Long Term Sigma selected, we will get a Long Term Capability Indices in the for of Cp, Cpl, Cpu, Cpk, which I think it's not quite right.  

2. If, however, we use the same data and run the capability study through the Control Chart Builder, we get a clear distinction between Within and Overall Sigma, using the correct indices terminology, i.e., Capability and Performance indices, respectively. 

Please note that the indices in point 1 above, in the form of Capability, are the same as the Performance indices in point 2.

Looking forward to read your comments on this as, at least for me, this generates confusion in the terminology used against its definitions.

Thanks,

Vasco Cachaco

[#2 · Accepted solution]

Hi Vasco,

Thank you for pointing this out. We've overhauled capability (and the distribution fitters) in the Distribution platform in JMP 15. So, this issue has been addressed and fixed starting with JMP 15 (which will be released in September).

Mia

[#3 · Accepted solution]

Prior to version 15, you can change preferences (file>preferences) so that the Distribution platform uses Pp/Ppk labelling instead of Cp/Cpk:

And in fairness, many of us have been using Cp for many years before it got rebranded as Pp!

[#4]

Hi Mia and David,

Thank you both for your replies, good to know that this issue is being addressed in JMP 15 !  :) 

Good tip on the Pp/Ppk labeling option too, through Preferences, it will help until September.

I wouldn't say that the Cp/Cpk have been 'rebranded' to Pp/Ppk though, but that the Capability definition has evolved with another level of 'resolution'. I have been exposed to this deferenciation since I started my journey in Six Sigma over 10 years ago and it always helped me and my project teams clearly understand these concepts and drive process improvements to the next level, i.e. towards getting processes closer to the Potential Capability (Cp/Cpk), with only common cause variation, when compared to the Actual Performance (Pp/Ppk), which is (most of the time) a lower number, when taking into consideration special cause variation. 

My point was that JMP has also taken this approach, which to me is the right thing to do, but it is (was) not being consistent with the terminology and therefore contradicting when displaying results, like in the example provided.

Thanks for your advice and help.

Regards,

Vasco

[#5]

Hello @VCO , It would probably be useful for the community to enumerate on the differences between Cpk and Ppk here in this thread.  It's nice that JMP has overhauled the labeling but the key to subverting confusion is understanding the calculations for what they are, and why they are.  Cpk and Ppk are in and of themselves structurally the same formula, except that Cpk uses a different estimate of sigma than does Ppk. How? 

In essence the sigma used to calculate Cpk is the root-mean-squared error of the sigma determined on the basis of rational (reasonable) subgroup assignment.  Said more simply, the sigma used here is in-effect the "average standard deviation" taken across the subgroups.  So Cpk considers meaningful subgroups in the context of the data.  A common meaningful subgroup is batch within lot or lot itself.  

Ppk on the other hand, does not make use of such a subgroup assignment. For this reason it uses the "overall" estimate of standard deviation in the calculation.  That overall estimate is just the "plain old" sample standard deviation that we know and love!  All the values are considered in relation to a common mean, X-bar.  So that mean is considered "shared" among all of the observations.  It's for this reason that when we calculate Ppk, it is said that we treat the data as "one homogeneous subgroup."  

Consider for example, these calculations the case of a one-sided lower specification limit (LSL):