Q&A: 2020 Questions, Part 2

Posted by Sten Westgard, MS

A recent question came in from Russia:

"We calculate CVa as long-term analytical CV from our routine internal QC. And here we've faced with two internal opinions regarding the CVa to take into account while calculating Sigma:

1. We should take in our CVa final calculation only those points of long-term internal QC, which are due to our analytical system in-control conditions. So we take only those points, which don't have any violations of our rejecting QC rules (for example, 1-3s).
2. We should take in our CVa final calculation absolutely all points of our long-term internal QC, including those which have violations with our rejecting QC rules and which are not accepted in routine IQC."
   

The answer, after the jump

The general approach is to make the standard deviation and mean out of data that is stable. If an error or outlier occurs, that is correctly excluded from the calculations because that does not represent the stable performance, and it represents a condition that must be corrected.

If you wish to determine the Sigma metric by the number of defects, you should adopt the counting approach, not the equation of estimation through observed variance and bias. The problem with that approach is that in many cases, you could be missing real errors or generating false rejections, so the table look up of Sigma metric will be impacted adversely.

In other words, don't try to include the observed outliers as part of the Sigma metric equation.

Comments

[Sofía]

What about taking into account the points between 3sd and 4sd? These are not strict outliers.

[DR PRASANTHI]

yes...Do not include the outliers in CVa calculation as performance cannot be judged based on outliers which indicate error in the system and inclusion of outliers leads to wide SD which would again affect the defined SD limits