Gage R&R calculator
Paste a crossed study — parts, operators, trials — and get the variance components, %study variation, and number of distinct categories the way Minitab’s ANOVA method computes them. Nothing is uploaded.
| Study | 10 parts × 3 operators × 2 trials |
|---|---|
| Total Gage R&R | 5.14% study variation · 12.89% tolerance |
| Distinct categories (ndc) | 27 |
| AIAG verdict | Acceptable (< 10% study variation) |
| Source | VarComp | % Contribution | Study var (6σ) | % Study var | % Tolerance |
|---|---|---|---|---|---|
| Total Gage R&R | 0.0001661 | 0.26 | 0.07733 | 5.14 | 12.89 |
| Repeatability | 0.0001162 | 0.18 | 0.06468 | 4.30 | 10.78 |
| Reproducibility | 0.00004989 | 0.08 | 0.04238 | 2.81 | 7.06 |
| Operator | 0.00004989 | 0.08 | 0.04238 | 2.81 | 7.06 |
| Part-to-part | 0.06282 | 99.74 | 1.504 | 99.87 | 250.63 |
| Total variation | 0.06298 | 100.00 | 1.506 | 100.00 | 250.97 |
ANOVA method; the part × operator interaction is pooled into repeatability when its p-value is above 0.25 (pooled for this data), as Minitab does.
Need nested designs, Type 1 Cg/Cgk, attribute agreement, or Gage R&R graphs? Open SpcLite →
What a Gage R&R study measures
A crossed Gage R&R splits the observed variation into the parts themselves and the measurement system. The measurement system has two pieces: repeatability, the spread when the same operator measures the same part again, and reproducibility, the difference between operators. If the gage takes up a large share of the variation, control charts and Cpk calculated from its readings are unreliable.
AIAG acceptance criteria
| % Study variation | Decision |
|---|---|
| Under 10% | Acceptable measurement system |
| 10% to 30% | May be acceptable depending on the application, gage cost, and repair cost |
| Over 30% | Unacceptable — improve the gage or the measurement method |
The number of distinct categories (ndc = 1.41 × σpart / σgage, rounded down) should be 5 or more. Use % tolerance instead of % study variation when the gage is used to sort good and bad parts rather than to monitor the process.
How to run the study
- Pick 10 parts that span the real process range, not 10 nearly identical parts.
- Use 2 or 3 operators who normally take the measurement, and 2 or 3 trials each.
- Randomize the measurement order and hide part numbers so operators cannot recall earlier readings.
- Every operator measures every part the same number of times (a balanced crossed design).
- For destructive tests where each part can only be measured once, use a nested study instead.
ANOVA vs Xbar-R method
The older average-and-range method estimates repeatability from ranges and reproducibility from operator averages. The ANOVA method used here gives the same quantities from variance components and also detects a part × operator interaction — operators who disagree only on some parts. It is the default in Minitab and preferred by the AIAG MSA manual.