Gage R&R (gage repeatability and reproducibility) is a study that measures how much of your observed variation comes from the measurement system itself rather than from the parts. It answers whether your gauge and operators can be trusted, usually reported as the percent of study variation the measurement consumes.
It matters because every number a plant acts on, a capability index, a control chart point, a pass/fail on an inspection, assumes the measurement is telling the truth. If a third of the variation you see is actually the gauge and the operators disagreeing with themselves, you are steering by a compass that wobbles. Gage R&R is how you find out before that noise costs you a scrapped-but-good batch or a shipped-but-bad one. This guide covers what the study measures, how to run one, and how to read and act on the result.
What is gage R&R?
Gage R&R is the core of measurement systems analysis (MSA): a designed study that splits your total observed variation into part-to-part variation (the real differences you want to see) and measurement-system variation (the noise you want to eliminate). The measurement noise is itself split into two parts, repeatability and reproducibility, which is where the name comes from. If the measurement system eats too much of the total variation, you cannot trust it to tell good parts from bad.
The idea sits underneath all data-driven quality work. The U.S. National Institute of Standards and Technology, in its engineering statistics handbook, frames measurement systems analysis as characterizing the measurement process so you know how much of the observed scatter is the process and how much is the gauge (NIST/SEMATECH e-Handbook, Measurement Process Characterization). Get that split wrong and every downstream statistic inherits the error.
What is the difference between repeatability and reproducibility?
Repeatability is the variation you get when one operator measures the same part multiple times with the same gauge. It is the gauge's own inconsistency, worn contacts, a sloppy fixture, a display that drifts. Reproducibility is the variation you get when different operators measure the same part with the same gauge. It is the disagreement between people, different technique, different pressure, different interpretation of where to take the reading.
The distinction matters because the fixes are different. High repeatability points at the equipment: recalibrate, replace, or redesign the fixture. High reproducibility points at the method: the operators are measuring differently, so you need clearer standard work for how to take the measurement, better training, or a fixture that removes the human judgment. A study that fails tells you not just that the measurement is bad, but which of the two to go fix first.
How do you run a gage R&R study?
A standard gage R&R study, done the AIAG way, uses about 10 parts, 3 operators, and 2 or 3 trials each. The parts should span the real range of what you produce, and the whole thing should mimic normal measuring conditions. Here is the procedure:
- Select the parts. Take about 10 parts that represent the actual spread of your process, not 10 good ones. The study needs to see real part-to-part variation to compare the measurement noise against.
- Pick the operators and gauge. Use 3 operators who normally run this measurement, and the actual gauge from the floor. Testing a pristine lab gauge tells you nothing about the one in use.
- Number the parts hidden from operators. Mark parts so the person measuring cannot see the number or remember a prior reading. Bias creeps in the moment an operator knows which part they are holding.
- Measure in random order. Have each operator measure all parts once in random order (one trial), then repeat for the second and third trials, re-randomizing each round so no one is anchoring to a recent number.
- Record every reading. Log all measurements against part and operator. Do not round, average on the fly, or discard outliers, the outliers are the signal.
- Calculate the variation. Use the ANOVA method (preferred) or the average-and-range method to split the total variation into part, repeatability, and reproducibility, then compute the %study variation and the number of distinct categories.
- Judge and act. Compare the result against the acceptance criteria below, and if it fails, fix the gauge or the method before you trust another measurement from it.
| Study design element | Typical value | Why |
|---|---|---|
| Parts | ~10, spanning the real range | Enough part variation to compare against |
| Operators | 3, the ones who really run it | Captures true reproducibility |
| Trials per operator | 2-3 | Captures repeatability |
| Order | Randomized, blind to part number | Removes memory and bias |
| Method | ANOVA (or average-and-range) | Splits the variation sources cleanly |
By the numbers. The widely used acceptance criteria come from the AIAG Measurement Systems Analysis (MSA) reference manual, 4th edition: a %study variation (%GRR) under 10% means the measurement system is acceptable; 10% to 30% is conditionally acceptable, allowed with justification for non-critical characteristics; and 30% or higher is unacceptable. Separately, the number of distinct categories (ndc) must be at least 5, a system can pass on %GRR and still fail on ndc, and you must meet both. NIST's engineering statistics handbook covers the underlying variance-component math these thresholds sit on (NIST/SEMATECH e-Handbook).
How do you read the %study variation result?
Read it as the share of your study's total variation that the measurement system is responsible for, and lower is better. Under 10%, the gauge is quiet enough that what you see is essentially the parts. Between 10% and 30%, the gauge is borderline: usable for non-critical features with a documented reason, but not something to rely on for a tight tolerance or a safety characteristic. At 30% or above, the measurement is consuming so much of the variation that you genuinely cannot separate good parts from bad, and any decision made with that gauge is a coin flip dressed up as data.
What do you do about a failing gauge?
First, read which component failed. If repeatability dominates, the problem is the equipment: recalibrate it, service the fixture, replace worn contacts, or upgrade to a gauge with finer resolution. A common quiet killer is resolution, if the gauge only reads to 0.001 and your tolerance is 0.004, the discreteness alone can sink the study, and no calibration fixes that; you need a better instrument.
If reproducibility dominates, the problem is the method and the people. Write and train a clear measurement standard, add a fixture or locating feature that removes the operator's judgment about where and how hard to measure, and re-run the study. Do not "fix" a failing gauge by widening the tolerance or averaging more readings to hide the noise, that buries the problem in the data instead of removing it. And do not keep shipping decisions on a red gauge in the meantime; segregate or re-measure with a trusted instrument until the study passes.
Why does gage R&R matter for capability and SPC?
Because capability and control charts both assume the measurement is clean, and gage R&R is the thing that proves it. A process capability study that reports a poor Cpk might be telling you the process is bad, or that the gauge is bad and inflating the observed spread. You cannot tell which until you know the measurement system is trustworthy. The same is true for control charts: measurement noise widens the limits, hides real signals, and triggers false alarms, so an out-of-control point might be the gauge, not the process. This is why measurement systems analysis comes before capability and statistical process control in any serious quality effort. It is also the reason a detection score in an FMEA is only as good as the measurement behind it, a control that depends on an untrustworthy gauge is not really a control.
Gage R&R is a point-in-time study, but the trust it establishes has to be maintained, and that is where live data helps. When measurement results, out-of-tolerance events, and re-measures are captured at the point of inspection instead of on a clipboard, a gauge that is starting to drift shows up as rising disagreement long before the next scheduled study, and you can confirm it on the floor during a gemba walk instead of discovering it in a month-end audit. That live feedback is what Harmony gives a plant, and it is part of the broader discipline of lean manufacturing: decisions are only as good as the data, and the data is only as good as the gauge. CLS made that shift, from measurements found the next morning to measurements visible during the shift, which is what keeps a passing gage R&R from quietly going stale.