Nested vs. Crossed Gage R&R Models for Destructive Tests

[PhilX09]

Hi,

I've been looking at the AIAG's white paper "non-replicable GRR Case study", and I am confused as to why it uses a nested ANOVA model. The study seems to be set up such that the assumption can be made that each appraiser measures the same 10 parts, hence surely a crossed ANOVA model would be more suitable?

Any thoughts on this? I'm trying to design some destructive GRR studies, and this is a bit of a stumbling block!

Thanks!
Phil

 

[Miner]

Re: Nested vs crossed models for destructive tests

A fully crossed study means that all combinations of parts and operators are tested. That is, all operators test all parts multiple times. Destructive studies mean that the parts tested (and destroyed) are nested within the operator that tested them and the operator can only test them one time.

All destructive (non-replicable) R&R studies are nested. Parts are nested within operators. If you try to analyze a non-replicable R&R study using a crossed ANOVA, you will obtain incorrect results.

Please read my blog entry on Non-replicable Gauge R&R studies for more information.

 

[PhilX09]

Re: Nested vs crossed models for destructive tests

Thanks Miner, and your MSA blog has been very useful.

I guess I was confused by the statement in the Minitab 14 help file:

"choosing between a crossed or nested Gage R&R Study for destructive testing depends on how your measurement process is set up. If all operators measure parts from each batch, then use Gage R&R Study (Crossed)"

Assuming this isn't correct, why go to the effort of making each appraiser measure parts from the same batch? I understand the importance of using consecutive parts to represent one part for repeat trials, but why for between operator trials?

Thanks for your help on this!

 

[Bev D]

Re: Nested vs crossed models for destructive tests

Same reason as for within...You want to reduce as much of the 'part to part' variation as possible within and between operators for non-replicable test method. This way you minimize the confounding of part to part variation with repeatability and reproducibility error. There are many approaches recommended by AIAG that are quite simply not best practice.

 

[RaiNt83]

Re: Nested vs crossed models for destructive tests

Sorry to barge in on this thread... Miner would be able provide some insight on how to go about a destructive GR&R for attribute data, eg pass/fail data without having a reference value.