Learn · updated 2026-10-21
Gauge R&R: is your measurement system good enough?
Before you trust a control chart or a Cpk, check the gauge. A gauge R&R study splits the variation you see into the parts themselves and the measurement system: repeatability (same person, same part, measured again) and reproducibility (different people measuring the same part).
The usual study
10 parts that span the real process range, 3 operators, each measuring every part 3 times in random order without seeing earlier results. That is 90 measurements. The ANOVA method then estimates a variance for each source.
Worked example: the AIAG study
This is the example from the AIAG Measurement Systems Analysis manual (4th edition), as printed by SPC for Excel, which Minitab also ships as sample data. LimitLine's ANOVA reproduces the published results:
| Source | % contrib. | % study var |
|---|---|---|
| Total gauge R&R | 7.76% | 27.86% |
| Repeatability | 3.39% | 18.42% |
| Reproducibility | 4.37% | 20.90% |
| Part-to-part | 92.24% | 96.04% |
The part × operator interaction was not significant (p = 0.974), so, as Minitab does by default when p is above 0.25, it is pooled into repeatability.
Reading the three numbers
| Measure | Here | Common guideline |
|---|---|---|
| % study variation (gauge σ ÷ total σ) | 27.9% | Under 10% good; 10–30% may be acceptable depending on the use; over 30% unacceptable (AIAG) |
| % contribution (gauge variance ÷ total variance) | 7.76% | Under 1% good; 1–9% may be acceptable; over 9% unacceptable (Minitab) |
| Number of distinct categories (ndc) | 4 | 5 or more (AIAG) |
Verdict for this study: marginal. The gauge uses about 28% of the observed spread and can only sort parts into 4 groups. It may be fine for checking parts against a wide tolerance, but not for fine process control. Most of the gauge error here comes from reproducibility: operators differ more than repeat readings do, so look at technique and training first.
% study variation is not % contribution: it is a ratio of standard deviations, so it doesn't add up to 100%. Here √7.76% ≈ 27.9%.
Sources
- Automotive Industry Action Group (AIAG), Measurement Systems Analysis Reference Manual, 4th edition (2010)
- SPC for Excel, “Three Methods to Analyze Gage R&R Studies” (the AIAG data and ANOVA results)