Repeatability Calculations like MSA

[#1]

I have a dataset of several thousand parts that were remeasured x10 in the same operating conditions.

So no cross terms or operator things like that.  I wanted to calculate repeatability with a formula and not have 100,000 different parts to look at in the MSA platform.  I can just do var/mean but is there more to it that EMP calculations would give me? Does anyone know how to just get the formulas to write it out in jsl?

I'm open to other recommendations on how to define the spec of this type of dataset.

Thanks,

[#2]

I do not understand your situation. Here are my thoughts (mostly questions):

1. Are you measuring the "several thousand parts" in the exact same location or is the within part variation confounded with measurement precision repeatability? Are you not interested in other measurement system elements (e.g., discrimination, reproducibility, stability, bias, accuracy)?

2. Why are you measuring the parts 10 times?  This is a subgroup size of 10 which seems excessive to get an estimate.

3. While you can easily calculate summary statistics for the subgroups, one of the key aspects of EMP is to look at the data graphically and assess the consistency of the within subgroup variation BEFORE any summary statistics are calculated.

4. I don't understand what you mean by "define the spec of this type of dataset"?  The spec should be related to customer requirements (theoretically form, fit and function).

[#3]

Sorry, I'm not sure what you are asking. And I'm not sure where the difficulty is in using the MSA platform (or Fit Model,  or Fit Y by X, or...). In addition to @statman 's thoughts...

"Repeatability", using language from MSA, is just residual error (i.e., RMSE from a one way ANOVA) and there is a formula...but it is based on the pooled variance which would require calculating 100,000 standard deviations. So, why not let MSA or Fit Model or Fit Y by X do it for you?  In MSA, just put in your "Y, Response" and "Part, Sample ID" then hit OK.

Now, if each part is its own column (so you have 100,000 columns) then you will have to stack the data first...

Sorry if I've misunderstood your issue.