Attribute R&R
Richard:
If you are dealing with a measurement (lenght), why consider it as an "attribute"??
MSAs for Variables are more helpful for understanding how good or bad is our measurement process. It may help you to have an idea for the next steps in productive process improvements...
Attributes can only help us to simply detect no-go parts...
But...
Assuming that usage of go no-go gage and 100% inspection is demanded by your customer, I hope this simple concepts help you to perform an attribute study:
- Choose at least 25 or 30 "perfect" and "defective" parts and ask each of the operators to inspect them randomly, at least three times each part. Remember: randomly, and in a way that resemble their routine work.
- Write down their decisions in a check sheet and after that, obtain the following:
- Repe: Hability of the appraiser to "repeat" his/her decisions (agree with him/herself). Calculated as (# agreements / # parts inspected) and commonly referred in percentage. 90% is usually considered as "acceptable", but this is a "rule of thumb".
-Repro: Hability of all the appraisers as a whole to "repeat" their decisions among them. Calculated as (# agreements among all appraisers / # parts inspected). (90% again is acceptable)
Since you already know the exact condition of each part, then you can determine:
- Concordance: (# parts that agree with the standard / # parts inspected). 90% again.
- False Alarms: (# parts classified as "defective" when in fact is perfect / # of perfect parts). 5% is acceptable (AIAG)
- Wrong Classficaction (# parts classified as "perfect" when in fact is defective / # of defective parts). 2% is acceptable (AIAG).
- Mixed: (# of parts that were classified inconsistently along the inspections / # parts inspected)
You can use Excel.
Forget about Kappa, is useful for ordinal categories (extremely bad, bad, good, extremely good).
Hope this helps!