Gage R&R: measuring whether your measurements mean anything

A gage repeatability and reproducibility study quantifies how much of the variation in a set of measurements comes from the measurement system itself rather than from the parts being measured. Repeatability is the variation when one person measures the same part repeatedly with the same instrument; reproducibility is the variation between different people measuring the same parts.

It matters because every decision downstream assumes the numbers are real. A process capability study, a control chart, a dispute with a customer over a rejected batch — all of them are arithmetic performed on measurements, and if a third of the spread comes from the gauge, the conclusions are about the gauge.

How a gage R&R study is run

• Select parts spanning the range of variation the process actually produces — not ten parts from the same batch, which understates the total variation and flatters the result.
• Use the operators who really do the job, not the two most careful people available.
• Each operator measures each part several times, in randomised order, without seeing previous readings.
• Analyse with the ANOVA method where possible; the older average-and-range approach is simpler but does not separate the operator-by-part interaction.
• Report the result as a percentage — of total variation, or of the tolerance, and state which, because the two answer different questions.

The percentage against tolerance is the practical one

Percentage of total study variation tells you whether the measurement system can distinguish parts from one another. Percentage of tolerance tells you whether it can decide conformance, which is usually the decision being made. A system can look acceptable against a wide process spread and be useless against a tight specification, and quoting the flattering figure is a common and consequential omission. State both, and be explicit about which drove the conclusion.

Widely used guidance treats under ten per cent as acceptable, ten to thirty as conditionally acceptable depending on the application and cost, and above thirty as unacceptable. Treat these as conventions to be agreed with your customer rather than as physical constants — what matters is whether the measurement system is adequate for the decision it supports.

Distinguish the two failures

A poor repeatability number points at the instrument or the fixturing: worn gauges, unstable setups, parts that move. A poor reproducibility number points at method: operators holding the part differently, taking the reading at a different point, interpreting an ambiguous instruction their own way. The remedies are different — one is metrology and maintenance, the other is work instructions and training — and averaging them into a single verdict hides which one you have.

Keeping the studies

Studies are repeated when the gauge, the operators or the process change, and customers in regulated supply chains ask for them by part characteristic. Ettex Sheets holds the data and the calculation so the result is auditable rather than a number pasted from a tool, Ettex Records keeps the completed studies per gauge and characteristic with their dates, and the instrument’s own traceability is covered in calibration certificate.

Being direct: this is a spreadsheet and records approach, not statistical software. Dedicated tools do the analysis properly and are worth having; what this covers is designing the study so the number means something, which no tool decides for you.