Cpk - How to calculate when the number of samples is less than 30 pieces?

[maghsoodi]

How is calculated When the number of samples is less than 30 CPK ?
Excel software to do this work?

[harrysons]

If you talk in a sample (units) size may be some one expert can support you
IMHO the CPK is pertinent by subgroup size, Sub group Frequency and the last is Number of supgroup and Generally the number of subgroups is 25 or more..to get the good test for stability,

Excel software to do this work?
excel able to this job.

[maghsoodi]

In Reply to Parent Post by harrysons

If you talk in a sample (units) size may be some one expert can support you
IMHO the CPK is pertinent by subgroup size, Sub group Frequency and the last is Number of supgroup and Generally the number of subgroups is 25 or more..to get the good test for stability,

Excel software to do this work?
excel able to this job.

stability,

My number for the CPK is 20 pieces. Not twenty-subgroup

[maghsoodi]

How is calculated CPK When the number of samples is less than 30 Pieces ?(My number for the CPK is 20 pieces. Not twenty-subgroup)
THANKS BABAK MAGHSOODIAN

[harry]

In Reply to Parent Post by maghsoodi

How is calculated CPK When the number of samples is less than 30 Pieces ?(My number for the CPK is 20 pieces. Not twenty-subgroup)
THANKS BABAK MAGHSOODIAN

Welcome to the Cove.

There is no need for a new thread as the original thread title had been edited to clearly reflect your situation - less than 30 pieces.

Weekends are typically slow so please be patient. The crowd will be back by Monday morning, US time.

[Jim Wynne]

In Reply to Parent Post by maghsoodi

How is calculated When the number of samples is less than 30 CPK ?
Excel software to do this work?

You can't derive a legitimate Cpk number from a sample of 20 pieces or 30 pieces or 100 pieces, for that matter.

Cpk assumes:

• A statistically stable process
• Rational subgrouping and chronological sampling while production is running
• A large enough number of measured subgroups to represent the expected output of the process over time.

Cpk is not a very meaningful index under the best of circumstances, but unless the conditions above are fulfilled it's pretty much a waste of time.

[Sam Lazzara]

Here is some info on Cpk.
http://www.itl.nist.gov/div898/handb...ion1/pmc16.htm

Having a stable process is a prerequisite and you cannot determine that from a single sample. Another important consideration is making sure your distribution is "normal" and your sample standard deviation (s) is a good estimate for the population standard deviation (sigma). Do not assume your distribution is normal/gaussian unless you test it, and a sample of 20 does not provide great power for that test.

Also note what is stated on the NIST link:
Most capability indices estimates are valid only if the sample size used is 'large enough'. Large enough is generally thought to be about 50 independent data values.

The math is the easy part.

[Darius]

Sorry for being a little late, have been bussy.... but this other tread is good for the matter.

Sample Size Does Matter

On that post, I try to say, you can calculate with almost any sample size but, the important thing is What is the worst case value for such estimate. Maybe many of us like to show good numbers, but being real is important, and most of us take the estimates as the true value, not just an estimate.

There are other issues far more important than sample size, as is stated on this tread, the normality is the real deal, and sample size does not assure the Gaussian Distribution.