How to Interpret Caliper GR&R (Gage R&R) Graphical Result

Micrometers are susceptible to a heavy or light touch and to angle of engagement depending on the feature geometry. There is also an angle of vision and estimating or rounding in needle spaces between marks for an analog micrometer. This doesn’t exist with a digital micrometer. There may also be some within piece variation that effecting results. The physical aspects cannot be forgotten in a statistical study.

Miner i look tomorrow your guide cause a little late here. I observed the way that operators performed the measurements. None of them use the Rachet of the micrometer either check before each sample if the micrometer goes to zero value..so i strongly believe that alla the variability of the measurments comes from these 2 reasons.

Now i conduct again the survey where i train them to use always the rachet and check before the value of mcrometer if is zero

Miner said:

You did not ask any questions about this, so I will give an overall assessment based solely on this graphic.
The gage is not suitable for SPC due to the high %Study Variation of 77%, and the redundant ndc of 1. The gage is marginally acceptable for use as an inspection gage (subject to the usual qualifications in the AIAG manual) since the %Tolerance is 15%. The major sources of measurement variation is in both Reproducibility and in Repeatability. You will probably find improving Reproducibility to be easier to address.

Click to expand...

Can we have the opposite result?
with ndc=4


The question is how we explained that?
How a measurement system is "more" capable for Process control but not for Product control?

Atul Khandekar said:

Look at selection of parts. Your max part measures 52.56, min is 52.52: that's a range of 0.04 against part tolerance of 0.6 ?

Click to expand...

Hello can you explain this a little more, i was not fully aware of what do you mean.
thank you

New to statistics said:

Can we have the opposite result?
with ndc=4
View attachment 30054

The question is how we explained that?
How a measurement system is "more" capable for Process control but not for Product control?

Click to expand...

It really is simple if you look at a plot of the data and think about it. Just relying on tables of statistical output is not sufficient and can be misleading. (Google “Anscombe’s quartet”). You need to do the work to understand what is happening; this is not just about mathematical formulas. Try writing down in words that do not use math terms what %study variation and %tolerance variation mean. Your answer lies there…

Since this is also a 17 year old thread - it is easier if you can start a new thread with your own data…

New to statistics said:

Can we have the opposite result?
with ndc=4
View attachment 30054

The question is how we explained that?
How a measurement system is "more" capable for Process control but not for Product control?

Click to expand...

This can occur when the variation of the parts selected is greater than the tolerance. There are two potential reasons:

1. The process is not capable (i.e., Cp < 1)
2. The parts were deliberately selected from extreme cases in order to obtain an acceptable %SV and do not reflect the actual process variation. In other words, the study was rigged (i.e., falsified)

Miner said:

This means that your gage must be able to measure to 0.001 or smaller to be adequate. From the R chart above, you are measuring to smaller increments than this, so the gage has adequate resolution

Click to expand...

You mean from the value 0.005579 (UCL) that is smaller than 0.001 ?

From Post #9 above (in context):

Miner said:

The minimum of 5 possible ranges in the R chart is a check for sufficient gage resolution. If the gage cannot resolve difference into more than 5 bins, it is inadequate. In the range chart above, the UCL is 0.005579. Divide this by 5 and you get 0.001. This means that your gage must be able to measure to 0.001 or smaller to be adequate.

Click to expand...

New to Statistics - you have clearly many questions to ask on the subject of GRR which is great, with respect its not clear that you have understood all that has been conveyed to you both in this thread and another you started on this forum.

Can I suggest that you start a new thread, base it on just one of your studies, give us the background as below:

1-Why you are carrying out the study in the first place, you want to learn but is this for a current process that requires improvement, have you had auditors ask you questions, is it a research study for you to learn etc etc etc

2-What are the parts you are measuring, how did you select them, straight from a production box, one from every hour of production etc etc

3-How did you arrange for the study to be undertaken- was the study performed in an administration office, factory shop floor, were the repeat measurements taken by each operator one after the other or was the test randomised, were all the operators present when the measurements were taken or was the study blind i.e. each operator couldn't see each others measurements.

4. How did you select the operators, were they all familiar with the process, are these the operators that will be used in production or is that not applicable here?

5. How did you convey to the operators about what the requirements of the study were, did you train them before hand, did you write a procedure for them to follow?


When you get to this point, post the Range and Average Charts and the raw data, then, based upon your current knowledge describe what them charts are telling you about your study, what you conclude and what you would do next.

Let the forum respond and then ask a question at a time to check your understanding.

I hope this is helpful, everyone here wants to help.


Thanks