The 2nd edition of PPAP says we are suppose to calculate "Preliminary Process Capability" in the form of Ppk, and yet the SPC Ref. Manual says Ppk is "Process Performance" while Cpk is "Process Capability" ….so what gives?

Also by the definition of Ppk and Cpk formula, Ppk value cannot be greater than Cpk since the "sigma" in Ppk includes both common and special causes of variation. And yet the acceptance for Ppk is supposedly 1.67 but for Cpk it's 1.33.

Anyone's got an insight to this?????

First, although some time ago, I found this in a search from 1997:

David McGan
Contributor posted 03 August 1997 11:08 AM
----------------------
I'm almost embarrassed to ask the question, but I've not been able to find a good, clear explanation of the real difference between the Ppk and Cpk indices. Ppk, I know, is usually specified for short-term study results, and uses the calculated standard deviation in its determination. Cpk, on the other hand, is used for long-term study results, and uses the estimated standard deviation in its calculation. But can anyone give me the statistical rationale for this?
-----------snippo-----------

I'm a generalist. Take a look at http://Elsmar.com/ubb/Forum10/HTML/000021.html and see if that helps. Also see http://Elsmar.com/ubb/Forum10/HTML/000024.html

If those don't help, come back to this thread, or post a reply in either (or both) of the threads.

Any comments, Don?

[This message has been edited by admin (edited 01 October 1999).]

Hmmm. Lets see. First, thanks for ‘lighting the bulb.’ I have been busy this past few weeks and had forgotten this was posted. My apologies for the delayed response.

First, as seen in the links Marc gave I have made my feelings about this whole Ppk thing known. I feel it is just a gimmick that the AIAG group uses to muddy the water. Just MHO.

Typically, Ppk uses sample standard deviation (s) while Cpk uses an unbiased estimate if sigma (sigma hat). The reason (I suppose) is that there may be occasions where this is more applicable. This is not true. Even with small sample sizes, an unbiased estimate of sigma can still be obtained.

As far as the 2nd edition of the PPAP manual goes, I ain’t gotta clue. As I have stated before, I do not do the QS thing.

Regards,
Don

------------------
Just the ramblings of an Old Wizard Warrior.

Cpk is the 6 sigma range of a process's inherent variation where Sigma is estimated by R bar (Range Average) / D2 for the Sample Size.

Ppk is the 6 sigma range of a process's total variation where Sigma is estimated by the samples standard deviation.

Cpk is an on-ging activity. Ppk is used as a "snapshot" of the process variation at any given point in time. Ppk is useful to determine if a process will be capable (Cpk) if process conditions maintain stability. It is also useful for determining if a continious improvement activity has had an effect on the process.

Daryl

First, although some time ago, I found this in a search from 1997:

David McGan
Contributor posted 03 August 1997 11:08 AM
----------------------
I'm almost embarrassed to ask the question, but I've not been able to find a good, clear explanation of the real difference between the Ppk and Cpk indices. Ppk, I know, is usually specified for short-term study results, and uses the calculated standard deviation in its determination. Cpk, on the other hand, is used for long-term study results, and uses the estimated standard deviation in its calculation. But can anyone give me the statistical rationale for this?
-----------snippo-----------

I'm a generalist. Take a look at http://Elsmar.com/ubb/Forum10/HTML/000021.html and see if that helps. Also see http://Elsmar.com/ubb/Forum10/HTML/000024.html

If those don't help, come back to this thread, or post a reply in either (or both) of the threads.

Any comments, Don?

[This message has been edited by admin (edited 01 October 1999).]

This is an interesting resurrection of some old threads.

First, it shows how much message boards have advanced since 1999.

Second, it shows that the topic of how to calculate and interpret Cpk has not advanced at all since 1999.

Third, it is humorous that Don Winton suggested reading Sun-Tsu's The Art of War to understand Cpk. I read the book imore than 10 years ago and I am positive it did not mention Cpk so Don must have been taking some literary license in referring to it. I asked for the history of Cpk in another thread but I believe we can narrow it down to happening between the death of Sun-Tsu (470 BC) and 1990 if that helps anyone remember. :lol:

Bill Pflanz

// From earlier threads....
Cpk is the 6 sigma range of a process's inherent variation where Sigma is estimated by R bar (Range Average) / D2 for the Sample Size.

Ppk is the 6 sigma range of a process's total variation where Sigma is estimated by the samples standard deviation.

Cpk is an on-ging activity. Ppk is used as a "snapshot" of the process variation at any given point in time. Ppk is useful to determine if a process will be capable (Cpk) if process conditions maintain stability. It is also useful for determining if a continious improvement activity has had an effect on the process.
//

Back to my problem!
When our customers want information of our process they always want a Cpk value, this value should it bee Ppk or Cpk?

The Ppk tells me how the process has been or lokked like? Or should I present the cpk value at "right" this time when we are processing our process, because this only take some of the near subgroups/samples depending how many soubgroups (50 or sometimes 25 subgroups deepending of the process) my calculations take to the Cpk calculation from the process?

Should the customer ask for Ppk value instead? Cpk value is more for us as a producer who can control the process?
Any suggestions?
Thanks!

Should the customer ask for Ppk value instead? Cpk value is more for us as a producer who can control the process?
Any suggestions?
Thanks!

The main issue of Cpk (against ppk) is altho both use the same data (or could use it), ppk's variation estimate (total variation) is affected with the outliers a lot more than it's conterpart ( Cpk uses within variation estimate).

There are other effects that affect Cpk that most of the SPC software don't take in account (like autocorelation, a shame...). If, and only "if", the process it self is not represented by the Cpk value, you may have such kind of problem. And too much variation on indicator could be because of the sample size, many take Cpk or ppk indicators as an absolute value without the sample size concern.

Maybe not the common way to put the things, but if ppk is a good indicator.. why the control limits are using the within variation estimate?. The within variation estimate for variation show us the process variation not the data variation, it's like with "normalizing" data, you can obtain estimates that with the raw data (all) you will need much more data points. :bonk:

Yes, it still amazes me that people don't yet understand the Cpk/Ppk thing. It is very clear in the literature how to calculate and use each one...although as time goes by tpp many practitioners have looked for the "cliff notes" version wich too often was written by soemone who really didn't understand Cpk and Ppk...and the AIAG shortcut approach makes it worse (they aren't wrong - just not clear on why they do what they do and so there is a lot of confusion). This is further compounded by too many people usign "Cpk" when they Mean Ppk...(everytime I've probed this - >50 separate organizations - I find that hte formula teh requestor wants is the formual for Ppk, they just call it Cpk...ARRGH!

But even with the above said, Cpk and Ppk are really just psuedo statistics: trying to reduce variation to a single number. Maybe it's voodoo statistics??

back in 1992 the journal of quality technology ran a full issue of capability index articles...then editorially stated that they woudl never accept another article on capability indexes because they weren't statistically appropriate. (Although I believe tehy have given on that statement since then). Perhaps we should follow their initial lead and pass citizen based referendums to outlaw the use of these indexes???

Perhaps we should follow their initial lead and pass citizen based referendums to outlaw the use of these indexes???

Where can I sign? :agree1:

By the way, I am teaching an MBA course on Operational Management this quarter. The textbook I am to teach from tells me that when doing SPC, you put the target value as the center line and calculate the control limits about that.

No wonder our MBA's in this country are so . . . well, hosed up. :bonk:

While I empathize with the desire to get rid of capability indices because they oversimplify compex processes to a single number, it begs the question as to what is a reasonable simplification.

Should we ban the calculation of averages, because it oversimplifies the central tendency to a single number?
Should we ban the calculation of st dev's, because it oversimplifies variation to a single number?

The opposite extreme would be to simply pass on the raw data and make the customer/boss/auditor draw their own conclusions.

Of course, both extremes are sometimes useful, and both extremes sometimes obscure interesting information. The challenge is to find a reasonable degree of simplification for the need at hand. Sometimes a single number, like Cpk or mu, is helpful. Sometimes it is almost valueless.

(The challenge of choosing the right index and performing the correct calculations is problematic, but it is a separate issue. )


Tim F

Of course, both extremes are sometimes useful, and both extremes sometimes obscure interesting information. The challenge is to find a reasonable degree of simplification for the need at hand. Sometimes a single number, like Cpk or mu, is helpful. Sometimes it is almost valueless.

(The challenge of choosing the right index and performing the correct calculations is problematic, but it is a separate issue. )


Tim F
OK, I agree that you probably can't do with the indeces. But choosing the right index and performing the correct calculations is not sufficient for gaining the proper information. If you never look at the control chart to do if the data are in control, or worse, don't know how to read a control chart or interpret what it means, you will still get the wrong answer.

People are required to have driver's licenses to drive a car, it is amazing that people figure they don't need a license to work with data. :whip:

People are required to have driver's licenses to drive a car, it is amazing that people figure they don't need a license to work with data. :whip: :agree1: :)

Perhaps I should change my signature to
Tim Folkerts: 007, license to calculate. :lmao:

:agree1: :)

Perhaps I should change my signature to
Tim Folkerts: 007, license to calculate. :lmao:

I love it. That would look good on a business card or resume!

Thanks for the answers!
Back to my problem, have I understand it right:
- That my customer is asking for wrong indexes? They call it Cpk when it should be Ppk?
/Niotusen

Thanks for the answers!
Back to my problem, have I understand it right:
- That my customer is asking for wrong indexes? They call it Cpk when it should be Ppk?
/Niotusen
they probably are. ask them what they want the data for, which formula to use...the answer to the reason for the data should tell you which to give them. I always default to the Ppk value taken over time (depending on how my process varies so that I have a representative sample) as it provides the most useful information (given that indexes are fairly value weak) for teh actual long term perfomance of a stable system.
Cpk can understate the full range of variation if the subgroups aren't rational

Yehhaa!
Now the problem begin?
My customer does not know how to calculate Ppk or Cpk? What is the different they say? Isn´t this great?

To the next question that com up:

What are the rules of Ppk then 1.67 or??
What should we follow on our processes?

Ppk over 1,67 or Cpk over 1,33?

Or must the both be under control at the sametime we produce??

My customer has no qlue and I ´m confusius? (Thanks this forum is to help)

Thanks!

Try to look at it this way. . .

A sprinter runs on a short groomed track. . . a cross country runner runs on country roads or streets. The Sprinter can average 14 MPH on his track. But put the Sprinter on country roads, and he may average 11 MPH.

The Sprinter has the potential to run 14 MPH. . . but with the variation experienced on the country roads (hills, ruts, turns, etc.), he would only be expected to average 11 MPH.

Ppk is Short Term Process Potential - Sprinter - Minimal Variation - 14 MPH - 1.67 Ppk minimum

Cpk is Long Term Process Capability - Long Distance - More Variation - 11 MPH - 1.33Cpk Minimum

I hope this analogy helps. This is the way I describe the difference. Same process. . . just looking at it under difference circumstances.