Re: Capability Study on Attribute Data? Visual Inspection
I beleive the last poster is confusing MSA with process capability of categorical data (aka attributes).
Of course you can calculate a Cpk (technically a Ppk value) from categorical data. Many customers actually insist on the Ppk value instead of the yield number. go figure
You simply go "backwards" from your defect rate to a Z score to the Ppk/Cpk value. Ppk = Z/3
Thre are 2 reasons we wantto know a Ppk score:
- what is projected yield
- what is spread of process (for stackup issues etc.)
only 1 really applies for categorical data.
The true complicating issue here is when you have a highly capable process and you have no defects for large volumes. I typically calculate the upper confidence interval for the exact binomial and use that value as my "yield" number. (yes it's a worst case estimate, but it provides context between a sampel size of 20 and 20,000) there are other accepted approaches and I can post them later if someone is interested...