How do you run an MSA on a go/no-go gauge?
A variable Gage R&R needs numbers to work on, and a go/no-go gauge only gives you pass or fail. So you measure agreement, not repeatability in millimetres. What an attribute MSA checks, and the known-standard parts it depends on.
Someone sets out to qualify a plug gauge the way they were taught, three operators, ten parts, three trials, and grinds to a halt at the first cell of the sheet. The gauge does not read 12.03. It reads go or no-go. There is no number to put in the box, so the whole variable Gage R&R template is useless here, and the instinct is to fudge the gauge into producing a reading it was never built to give. The gauge is fine. The study is the wrong one. A go/no-go gauge needs an attribute MSA, and it asks a different question than the one Gage R&R answers.
Variable Gage R&R needs numbers a go/no-go gauge does not have
A variable Gage R&R works by splitting the total variation in a set of measurements into part-to-part variation and measurement-system variation, and to split variation you need variation, which means real numbers on a continuous scale. A go/no-go gauge, a plug, a snap gauge, a functional check, a visual pass-fail against a limit sample, produces exactly one bit of information per check. In or out. Good or bad. There is no spread to decompose, so there is nothing for the Gage R&R maths to work on. Forcing a pass-fail into a number, scoring good as one and bad as zero and running Gage R&R on that, does not rescue the study, it produces a meaningless index dressed up as a real one. The attribute gauge has to be assessed on its own terms.
You are measuring agreement, not repeatability in millimetres
Because the gauge only ever says good or bad, the thing you can actually test is how consistently it says the right thing. That is agreement, and an attribute MSA looks at it three ways. Within an appraiser, does the same person get the same call when they check the same part more than once, which is the attribute version of repeatability. Between appraisers, do different people checking the same part reach the same call, which is reproducibility. And against the standard, do their calls match the part's true, known condition, because two appraisers can agree perfectly with each other and still both be wrong. The statistic that puts a number on this is usually a kappa, which measures how much of the agreement is real and how much is just what you would expect from chance. Strong agreement between people but poor agreement with the standard means the gauge or the method has a built-in bias everyone has learned to repeat.
What an attribute study needs to be worth running
- Parts with a known, independently measured true condition, so you can score the calls against truth and not just against each other. Without known standards you are measuring consensus, not accuracy.
- A deliberate spread of parts, clearly good, clearly bad, and a set that sits close to the specification limit, because the middle is where an attribute gauge earns its keep or fails.
- Enough parts that the borderline group is not one or two lucky examples. A common working set is around fifty parts weighted toward the limit, checked by two or three appraisers, twice each.
- The appraisers kept blind to each other and to the previous result, so a call is a fresh judgement and not a memory of last time.
- The gauge and method recorded, the same reference data an inspection relies on, so a study result is tied to the exact gauge it qualified.
The boundary is the whole test
The mistake that guts most attribute studies is loading them with easy parts. A drawer of obviously good and obviously scrap pieces produces beautiful agreement and proves almost nothing, because no gauge struggles with the easy ones. An attribute gauge lives or dies at the decision boundary, the parts just inside and just outside the limit, and a study that does not deliberately include borderline parts has not tested the thing that actually goes wrong on the floor. This is also the hardest part of the work, because you need parts whose true condition near the limit is known precisely enough to argue with, which usually means measuring them on a better variable instrument first. Get the borderline set right and the study tells you something real. Skip it and you have a certificate, not an answer.
The free Gage R&R calculator runs the variable study, the numeric three-operator, multiple-trial kind, and it is the right tool the moment your gauge gives you a reading. It will not run an attribute study, because attribute agreement is a different calculation on go/no-go data, and pretending otherwise would be the exact fudge this article warns against. In VoraControl, gauges and methods are held as the reference data capture runs against, so a gauge that has been qualified, however it was qualified, is the same record the floor inspects with. The study qualifies the gauge. The gauge record carries that qualification into the everyday work.
Released control plans, inspection capture and SPC trends in one governed flow.
Talk to someone who understands manufacturing control and audit evidence. We do not do generic demos.