Gauge to Gauge (Gage) Studies

Good morning all,
Looking through the threads for testing the viabilty of a replacement measuring system has proved very helpful to a point.
I have concluded that a Youden plot and a Bland Altman plot seem to be the best tools for the job.
Was wondering if I might get some confirmation on my analysis of the two methods.

Please reference the attachments below

On the Bland Altman, there is only 1 point outside the limits of agreement, yet the dispersion of the data is wider than I would like to see.
Is the fact that is is somewhat equally distributed between upper and lower an indication that this would be a decent replacement?
Have done research on interpreting this method and still feel that I'm missing something in the analysis.

On the Youden plot, how much deviation from the 45 degree line is acceptable? I do understand that the interpretation for these are clinical more than a black and white statistical inference so I would like some second opinions / any advice that might help draw conclusions.
Same as above, I have researched this and still feel like I have not come to the concrete determinations that I would like.

Find the cove to be an invaluable resource, spend alot of time on here finding answers, any help on this particular issue would be greatly appreciated.

Re: Gauge to gauge studies

Overall, you have made a correct assessment. Regarding the point outside the limits of the Bland-Altman plot, these limits are 95% confidence (+/- 2sigma), so a single point is of little concern.

You did not specify what you are measuring, so I do not know whether the range of parts is sufficiently wide. It should be as wide as the anticipated measurement range. I would definitely try to find additional samples at each end of the range.

Re: Gauge to gauge studies

Miner said:

Overall, you have made a correct assessment. Regarding the point outside the limits of the Bland-Altman plot, these limits are 95% confidence (+/- 2sigma), so a single point is of little concern.

You did not specify what you are measuring, so I do not know whether the range of parts is sufficiently wide. It should be as wide as the anticipated measurement range. I would definitely try to find additional samples at each end of the range.

Click to expand...

For the sake of this particular sample set, the spec range would be 3.610 +/- .010.
So the samples do emcompass the complete range.
Is the fact that the data in the Blandalt show a range of +/- .005 disconcerting?

On to the Youden plot;
How does one determine how far from the thoretical 45 degree line your data can be dispersed and still be deemed acceptable?

Thanks for your time and help, very much appreciated.

Re: Gauge to gauge studies

See this linked thread Two gauges - Two measurements correlation I do not know how to explain, particularly post #2 from Bev D.

Another interesting read is the following from MedCalc.

Re: Gauge to gauge studies

" No trend or “CV” effect can be observed over this range of sample results. The differences between measurement methods appears to be random, and so the calculated 2 σd limits can be considered a good estimation for future results. "


The above quote is from the MSA doc from the link.This statement fits my data when plotted also. I guess I'm still confused as to whether the "estimation for future results" refers to the fact that the gauge may be that far off in it's accuracy (in my plot that would mean a +/- .005" range)?


Thanks again for the reply, Had not seen the MSA sheet. very well written, there just seems to be one small part that I am not grasping.

Re: Gauge to gauge studies

MEthod comparison is lookign for two things: bias and repeatability.

Both the Youden and the B-A plot show that the difference between your two systems is random: there is no bias. That's a good thing.

However, there is some repeatability error that is concernign to you.

YES, if you don't change anything then you can reasonably expect to see roughly a MAXIMUM of + .005 measurement error between the two systems.

Only YOU know if this is acceptable.

Remember you've only looked at the two systems in relationship to each other - you have not presented any data on the repeatability of the two systems themselves...If you feel that this is too much you will need to go back and perform the repeatability study on each individual system: this may tell you that one is more repeatable than the other.

If we may ask: what feature are you measureing? this may enable us to give better guidance on the acceptability of the measurement error.

Re: Gauge to gauge studies

Bev D said:

MEthod comparison is lookign for two things: bias and repeatability.

Both the Youden and the B-A plot show that the difference between your two systems is random: there is no bias. That's a good thing.

However, there is some repeatability error that is concernign to you.

YES, if you don't change anything then you can reasonably expect to see roughly a MAXIMUM of + .005 measurement error :mybad:between the two systems.

Only YOU know if this is acceptable.

Remember you've only looked at the two systems in relationship to each other - you have not presented any data on the repeatability of the two systems themselves...If you feel that this is too much you will need to go back and perform the repeatability study on each individual system: this may tell you that one is more repeatable than the other.

If we may ask: what feature are you measureing? this may enable us to give better guidance on the acceptability of the measurement error.

Click to expand...

That was Exactly what I was looking for!
One gauge is definitely more repeatable than the other with a total R&R of about 4.9%.


Thanks to you and Miner, I find both of you to be very helpful in your postings.

Re: Gauge to gauge studies

Bev D said:

MEthod comparison is lookign for two things: bias and repeatability.

Both the Youden and the B-A plot show that the difference between your two systems is random: there is no bias. That's a good thing.

However, there is some repeatability error that is concernign to you.

YES, if you don't change anything then you can reasonably expect to see roughly a MAXIMUM of + .005 measurement error between the two systems.

Only YOU know if this is acceptable.

Remember you've only looked at the two systems in relationship to each other - you have not presented any data on the repeatability of the two systems themselves...If you feel that this is too much you will need to go back and perform the repeatability study on each individual system: this may tell you that one is more repeatable than the other.

If we may ask: what feature are you measureing? this may enable us to give better guidance on the acceptability of the measurement error.

Click to expand...

A diameter of a plastic part.

Re: Gauge to gauge studies

Brian P said:

A diameter of a plastic part.

Click to expand...

Just to be sure - did you mark the part so that the operators measured the exact same spot (diameter) on each trial?

Re: Gauge to gauge studies

Yes, there was training and observation of methods involved pre-study.
1 of the methods involve multi-point meas. with average being used as end result.