How to use the gage R&R calculator
- Run a crossed study. Select parts that span the real process range (typically 10), have each appraiser (typically 3) measure every part 2 or 3 times in random order, without seeing earlier readings.
- Enter the data in long format — one row per measurement with the columns Part, Operator, Measurement. Paste from Excel, upload a .csv/.xlsx file, or start from the template. The number of trials is inferred from the repeats.
- Add the tolerance (USL − LSL) to get %Tolerance (the P/T ratio) and, if you know it, the historical process standard deviation to get %Process.
- Read the verdict, the ANOVA and variance-component tables, and the charts. Print the report or save it as a PDF.
The study must be balanced: every operator measures every part the same number of times. If it is not, the tool shows a count table with the offending cells highlighted.
What gage R&R measures
Every measurement is the true value of the part plus measurement error. A gage repeatability and reproducibility study splits the observed variation into:
- Repeatability (EV, equipment variation) — scatter when the same operator measures the same part again with the same gage.
- Reproducibility (AV, appraiser variation) — differences between operators, including the operator × part interaction (an operator who reads some parts differently).
- Part-to-part variation (PV) — the real differences between the parts.
Gage R&R is the combination σ²GRR = σ²EV + σ²AV, and total variation is σ²TV = σ²GRR + σ²PV.
The ANOVA method
The calculator fits the two-way random-effects model with interaction and reports DF, SS, MS, F and p for Part, Operator, Part × Operator and Repeatability. With p parts, o operators and r trials the variance components are:
| Component | Interaction kept | Interaction pooled |
|---|---|---|
| Repeatability | MSrep | MSpooled |
| Operator × Part | (MSP×O − MSrep) ÷ r | — |
| Operator | (MSO − MSP×O) ÷ (p·r) | (MSO − MSpooled) ÷ (p·r) |
| Part-to-part | (MSP − MSP×O) ÷ (o·r) | (MSP − MSpooled) ÷ (o·r) |
When the interaction is not significant (p above the α you set, 0.05 by default) its sum of squares is pooled into repeatability and the reduced model is used, which is the convention in the AIAG MSA manual and in statistical packages. Negative variance estimates are set to zero and flagged. The F-distribution p-values are computed exactly from the regularized incomplete beta function, not looked up in a table.
The average & range method
The classic AIAG form estimates the same quantities from ranges: EV = R̄ × K1, AV = √[(X̄diff × K2)² − EV² ÷ (n·r)], GRR = √(EV² + AV²), PV = Rp × K3 and TV = √(GRR² + PV²). R̄ is the average range of the trial readings, X̄diff the difference between the highest and lowest appraiser averages and Rp the range of the part averages. The K factors are 1 ÷ d2*; the tool uses the published values (K1 = 0.8862 and 0.5908 for 2 and 3 trials, K2 = 0.7071 and 0.5231 for 2 and 3 appraisers, K3 for 2 to 10 parts) and derives the factor from d2* for other study sizes, stating when it does so. The range method cannot separate the interaction, so prefer the ANOVA numbers when the two disagree.
How to interpret %GRR and ndc
| %GRR | Usual decision |
|---|---|
| Under 10% | Acceptable measurement system. |
| 10% to 30% | Conditionally acceptable, depending on the importance of the characteristic and the cost of improvement. |
| Over 30% | Not acceptable — improve the measurement system. |
- %Study Var = 100 × σGRR ÷ σTV compares the gage with the variation seen in the study. Use it when the gage is used for process control — and only if the parts represent the process spread.
- %Tolerance = 100 × 6σGRR ÷ (USL − LSL) compares the gage with the specification. Use it when the gage sorts good from bad parts.
- %Process uses a historical process σ instead of the parts in the study.
- %Contribution is the ratio of variances (σ²GRR ÷ σ²TV); 10% and 30% of study variation correspond to 1% and 9% contribution.
- ndc = ⌊1.41 × σPV ÷ σGRR⌋, the number of distinct categories the system can resolve. It should be 5 or more.
The X̄ chart by operator uses limits based on repeatability only, so about half or more of the points should fall outside the limits; the R chart should be in control. Once the gage is proven, continue with a process capability study or an X̄–R control chart.
Worked example
The sample data set has 10 shaft diameters measured 3 times each by 3 operators (90 readings) with a tolerance of 0.25 mm. Load it to follow the numbers.
Interaction: F = MSP×O ÷ MSrep = 0.00003086 ÷ 0.00002858 = 1.08, p = 0.39 > 0.05, so it is pooled: MSpooled = (0.00055553 + 0.00171467) ÷ (18 + 60) = 0.00002911. This is the repeatability variance.
Operator: σ²O = (0.0012159 − 0.0000291) ÷ (10 × 3) = 0.00003956. Part: σ²P = (0.0049860 − 0.0000291) ÷ (3 × 3) = 0.00055077.
Gage R&R: σ²GRR = 0.00002911 + 0.00003956 = 0.00006867, σGRR = 0.00829 mm; σTV = √(0.00006867 + 0.00055077) = 0.02489 mm.
%GRR = 100 × 0.00829 ÷ 0.02489 = 33.3% of study variation (not acceptable), 6 × 0.00829 ÷ 0.25 = 19.9% of tolerance (conditionally acceptable), and ndc = ⌊1.41 × 0.02347 ÷ 0.00829⌋ = ⌊3.99⌋ = 3. The gage is marginal for sorting parts and too coarse for process control; reproducibility (operator differences) is the larger component.
Assumptions and limitations
- Crossed design: the same parts are measured by every operator. Destructive tests need a nested design, which this tool does not fit.
- Balanced data with at least 2 parts, 2 operators and 2 trials; 10 × 3 × 3 or 10 × 3 × 2 is the common practice. Few parts or operators give very wide uncertainty on the variance components.
- Measurement error is assumed roughly normal and constant over the measuring range; bias, linearity and stability are separate studies.
- %Tolerance needs a two-sided tolerance. For a one-sided specification judge by %Study Var or %Process.
The method follows the AIAG Measurement Systems Analysis reference manual (4th edition) and Montgomery, Introduction to Statistical Quality Control. If the measurement distribution itself is in question, check it with the normality test.
Frequently asked questions
What is an acceptable gage R&R percentage?
By the usual AIAG guideline, a %GRR below 10% is acceptable, 10% to 30% is conditionally acceptable depending on the application and cost, and above 30% is unacceptable. The number of distinct categories (ndc) should also be 5 or more.
What is the difference between the ANOVA method and the average and range method?
Both estimate repeatability, reproducibility and part variation. The average and range method uses ranges and tabulated K constants and cannot detect an operator-by-part interaction. The ANOVA method uses mean squares, separates the interaction and is statistically more efficient, so it is preferred when software is available.
How many parts, operators and trials do I need for a gage R&R study?
The common design is 10 parts, 3 operators and 3 trials (90 measurements), or 10 parts, 3 operators and 2 trials. The parts must represent the full range of the process, and every operator must measure every part the same number of times.
What is ndc (number of distinct categories)?
ndc equals 1.41 times the part standard deviation divided by the gage R&R standard deviation, truncated to an integer. It is the number of non-overlapping groups of parts the measurement system can reliably distinguish. A value of 5 or more is required for process analysis.
Should I use %study variation or %tolerance?
Use %tolerance (the precision-to-tolerance ratio) when the gage is used to accept or reject parts against a specification. Use %study variation or %process when the gage is used for process control and improvement, because there it must resolve the process variation rather than the tolerance.
Why is the operator-by-part interaction pooled into repeatability?
If the F test for the interaction is not significant (p greater than the chosen alpha, commonly 0.05), the data give no evidence of an interaction, so its sum of squares and degrees of freedom are combined with the repeatability error. This gives a better error estimate with more degrees of freedom.