ANOVA Reproducibility Error Contribution - When to recalculate/disregard variance

[QualityEngBrady]

Hello,

For my study I have an ANOVA Table that shows both OP (.099) and Part*OP (0.671) to be greater than the .05 Alpha requirement. I presume this means that the operator factor (Reproducibility) should have a 0% contribution to the overall GRR %StudyVar, however my Gage Valuation table still calculates the Reproducibility to be 0.76%.

Am I wrong about the 0% defacto contribution or what is the reasoning why I would still have a reproducibility error.

 

[Bev D]

I’m sure @Miner will provide a more detailed answer…for now there a few things to understand:
1) There will always be some mathematical difference. The question is whether or not the difference is real or simply random chance.
2) the p value is an indication of whether or not the mathematical difference is real or random chance.
3) There will always be some operator reproducibility ‘error’. This is not error in the traditional sense but it is a measure of the contribution of operator differences separate from the system variation (repeatability).

The two p values greater than .05 or 5% indicate that the operator differences are simply random variation not a ’real’ systemic difference.
The Operator reproducibility of 0.76% is INCREDIBLY small - this is a great result. I have seen operator reproducibility be the controlling factor in measurement variation upwards of 90+%. I have always been able to fix it and reduce to near zero - at least to a level that had no substantial effect on measurement variation…

There is no ‘defacto contribution’ of 0%. Whoever told you that or whoever wrote it is completely wrong.

 

[QualityEngBrady]

Hello @Bev D ,

Thank you for your response. This question arose from a Gage R&R with a Keyence LMX (Optical and Laser system) that determines the points of measurements from a pre-programmed profile and I would classify as a Automatic Gage. There is zero human interaction except for cleaning the stage and laying down the parts.

Do to the knowledge of how this system works to determine measurement data, I believe that the human operator has no significant impact on how the part measures I would like to completely negate the reproducibility contribution. Is that an OK assumption to make?

 

[Bev D]

In this case yes - there is no real operator contribution. But not because of ‘how the system works’. It is because the reproducibility variation in your study was at 0.76%. You need to understand this.

BUT, never think that a ‘high tech’ system has no operator contribution. Always do the test. Then you will know. Having dealt with simple and highly complex expensive measurement systems for over 40 years (including the Keyence and others that you have no idea about) I can say that the operator often has serious and large effects on even complex systems. And yes I’ve seen a few tests using the Keyence that have operator effects due to sample prep and placement…

Your knowledge and understanding of MSA is very meager (no offense this is normal and you are to be commended for asking questions).
I suggest you read Miner’s Blogs, My MSA presentation and check out the MSA resources in my “Essential References for Quality Engineering”

 

[Miner]

I’m sure @Miner will provide a more detailed answer…for now there a few things to understand:

@Bev D No need to add detail. You nailed it.

 

[QualityEngBrady]

Hello @Bev D ,

I will not take offense to that statement - but essentially it comes down to is complete a normal GRR with a standard test setup (3x3x10,2x3x25) and interpret the Operator and OperatorXPart calculated statistics to determine if the measurement system is Operator (including part setup) dependent. Never exclude the Reproducibility term from calculations as the math will prove it out.

My only outstanding questions is why are terms excluded if the P value is >.25 (non-interaction) instead of 0.05 alpha?