Combined GR&R

[#1]

Hi everyone,

I'm conducting a GaugeR&R Analysis using a crossed 6-sigma process,

Where my setup is 5 units, 3 employees and 3 stations. 

How can I tell what are the variance components when I calculate a combined setup where I aim for a unified single result?

(using JMP 18)

Thank you in advance

[#2]

You'll need to provide some more detailed information regarding the exact structure of the experiment you conducted and then some additional context. For example, is the response numeric, ordinal or categorical/attribute. Nesting? Crossing? You mention 'crossed'...but where in the structure? How many replicate measurements. And the model?

I'm not a big fan of classic AIAG Gage R&R methods (if that's where you are headed) for lots of reasons I won't go into here...especially when it comes to variance components analysis. It's possible to do this type of analysis in JMP...but JMP clearly steers you to Don Wheeler's EMP approach...so once you have your experiment conducted and responses collected I suggest starting here in the JMP documentation for a sound EMP based analysis workflow:

EMP in JMP  

[#3]

IMHO, the problems with Gage R&R are less to do with the analysis and more to do with the data collection strategy.  Issues include:

1. Gage R&R is biased to the measurement system components (multiple layers are components of measurement system sources; e.g., gage and operator with one layer for sample variation for comparison)
2. Folks don't recognize the importance of selecting samples for the study.  What variation do the samples represent?  Since you are comparing the measurement variation to the sample variation, if you choose samples that are similar (e.g., 5 consecutive samples) vs. samples that may vary wildly (e.g., random samples) you can arrive at completely different conclusions about the capability of the measurement system.
3. Unfortunately, folks tend to do gage R&R and "christen" the measurement system as good forever. Studies where results are applied beyond the inference space.
4. Mixed model (crossed and nested), are difficult to assess stability of the measurement system.  Control charts are not appropriate for crossed studies.

Don Wheeler has written some good papers on the subject:

Wheeler, Don (2006) “An Honest Gauge R&R Study”, 2006 ASQ/ASA Fall Technical Conference, No.189

Wheeler, Donald (2020) “Gauge R&R Methods Compared: How do the ANOVA, AIAG, and EMP approaches differ?”, ASQ Statistics Division Newsletter, Vol.39, No.1

[#4]

This is the ANOVA result in the crossed GRR validation.

GRR uses the value of the variance component and the value of the variance component is from ANOVA.