Discussion in 'DOE - Design of Experiments' started by _Zeno_, Oct 15, 2015.
How does GRR affect DOE design and interpretation?
A GRR should be performed prior to designing an experiment to ensure that you can trust your measurement system. If the measurement system has excessive variation, it could mask the very effects that you want to see in the experiment by inflating the Residual Error portion of the ANOVA table. If you properly calculate the number of replicates in advance, measurement variation could require a larger number of replicates than would otherwise be necessary. At a minimum, you would need to run repeats and average them.
It is better to have a good measurement device with minimal variation.
Interpretation: A large amount of measurement variation will be seen as part variation (aka 'noise') in either graphical or statistical analysis of the DoE results. This could result in missing true differences in either main effects or interactions.
Design: If you have minimal measurement variation then you can get by with a smaller sample size for each trial. If you have substantial measurement variation taht you cannot easily reduce, you will need to increase your sample size for each trial.
I recommend checking out the following in the Resources section of this forum. (Resources tab is on the upper left hand area next to the forums tab)
MSA for Functional Tests: page 6 and 7 explain an MSA with a large amount of measurement variation and a compensation scheme.
Profound Statistical Concepts Part 2 Example 1: This example shows a classical DoE (3 factors, 2 levels) with horrible measurement error.
You might also find these useful:
The Statistical Foundation Cracks in the Popular Gauge R&R Method
MSA Verification and Validation
Separate names with a comma.