Developer Tutorial: Designing and Evaluating Measurement Systems Studies Prior to Implementation
calking
| Updated on
Note: All the capabilities shown are in JMP 17, although JMP Pro 17 is used for the demo. The video includes Q&A and comments from the presenter Caleb King @calking about how MSA works plus plans for possible enhacements.
See how to:
- Understand basics of MSA Design
-
MSA primary focus is on decomposing variation in product measurements into distinct sources, with the goal to reduce or elimnate one or more sources
-
The metric of interest is not an effect size (i.e. difference in mean behavior), but rather the variance contribution from a particular source
-
Underlying models are primarily random effects models or even mixed effects models if biasing terms are included
-
Nested factors are also common in MSA, but not exclusively
-
-
See improvements made in JMP 17
-
Better diagnostics that more closely align with MSA analytics platforms
-
Greater user interaction that allows exploration of what-if scenarios
-
Fexible Design structure interface so users can specify nesting/crossing relationships in the design
-
- Understand that diagnostics are based on simulation of variance estimates, based on the standard ANOVA approach
-
Example is a 3- factor fully crossed MSA where mean squares have known Chi-square distributions that are easily be simulated and then used to compute confidence intervals, variance proportions, etc.
-
- Include nested and crossed structures
- Make design and evaluate diagnostics for two examples
The video includes Questions ansered at the webinar by Caleb @calking, Don McCormack @DonMcCormack and Jerry Fish @JerryFish.
One person asked online if nesting would occur when testing same methods on different equipment. They responded: Probably not, but it would ultimately depend on how the experiment was run what assumptions were made about the methods and equipment. If you considered the equipment/method to be a random observation from a much larger population, then they may be considered nested (in some way). If you considered them from a fixed set of values (e.g., types of equipment), then they would more likely be treated as crossed.
Resources
- Caleb's blog and interview
- More about Measurement Systems Analysis in JMP
- More about using EMP in JMP
