Gauge R&R calculator: ANOVA and average & range, free and validated

A gauge R&R calculator splits the variation in your study into repeatability, reproducibility and part-to-part variation, and tells you how much of it is the measurement system. Paste your readings from Excel, choose ANOVA or average and range, add the tolerance if you have one, and you get %GRR, %Contribution, %Tolerance, the number of distinct categories and the ANOVA table, with the verdict under the AIAG criteria.

The calculations run in your browser and nothing is sent anywhere. The calculator opens with the AIAG example data and reproduces the values published for it by Minitab (ANOVA) and SPC for Excel (average and range), to the last published decimal (worked example). New to gauge studies? Read the gauge R&R guide first.

How to use the calculator

1. Start from the example. The calculator opens with the data of the AIAG MSA example: 10 parts, 3 operators, 3 trials, tolerance 8. Look at the results first, so you know what the output means. Clear the data empties the table.
2. Set the study size. Choose the number of operators (2–5), trials per part (2–5) and parts (2–15). Values already entered stay in place; making the study smaller drops the values that no longer fit.
3. Enter or paste the readings. Copy the block of numbers from Excel or Google Sheets and paste it into the first cell, or use Paste a whole table, which also skips a header row and a column of part labels. Empty or invalid cells are listed in red.
4. Add the tolerance (optional). Enter USL − LSL, in the same units as the readings, to get %Tolerance. For a one-sided specification, leave it blank and judge the gauge on %Study Variation.
5. Choose the method. ANOVA is the default. Options lets you switch study variation to 5.15 × SD and keep the operator × part interaction in every study, to match older material or other software.
6. Read the result. The panel on the right gives %GRR of study variation, %Contribution, %Tolerance and ndc, and applies the AIAG bands. Below the table you get every component, a chart, the ANOVA table or the average and range steps, both methods side by side and a CSV download for your records.

How to lay out your data

The calculator expects a crossed study, the common design: every operator measures every part, the same number of times, in a new random order each round (SPC for Excel, 2015; Minitab example).

Put one row per part, and the columns grouped by operator, then trial: A1, A2, A3, B1, B2, B3 and so on. This is the first part of the AIAG example in that layout:

Numbers use the US format: a dot for decimals and, optionally, commas for thousands (1,234.5). A value like 0,25 is flagged rather than guessed. Tab-separated text (what spreadsheets copy), CSV, semicolons and spaces are all accepted.

How the calculator works

The gauge R&R guide explains every step with the same example. In short:

ANOVA fits a two-factor random model, part and operator, with their interaction. The variance components come from the mean squares, and any component that comes out negative is set to zero (Minitab, methods and formulas; NIST e-Handbook 2.4.4). The operator × part interaction is tested with F = MS_interaction / MS_{repeatability}: if its p-value is above 0.05, it is pooled into repeatability and the model is refitted without it, as in the published Minitab example. The 0.05 cutoff is a software setting, not a universal rule; some published analyses keep the interaction even when it is not significant (SPC for Excel, ANOVA Gage R&R Part 1), and Options lets you do the same.

The average and range method uses ranges instead of sums of squares (SPC for Excel, 2015; Ermer, 2006):

• EV = R̿ × K1, from the average range of each operator-part cell;
• AV = √((X̄_diff × K2)² − EV² / (parts × trials)), from the range of the operator averages;
• PV = R_p × K3, from the range of the part averages;
• GRR = √(EV² + AV²) and TV = √(GRR² + PV²).

K1 = 1/d2 and K2, K3 = 1/d2* come from the statistics of the range of normal samples, which is why they change with the number of trials, operators and parts (Ermer, 2006). The calculator computes them by numerical integration for any study size instead of copying a table; for 3 trials, 3 operators and 10 parts it gives 0.5908, 0.5231 and 0.3146, the values in the AIAG manual as reproduced by SPC for Excel (2015).

Both methods then report each component as a share of total variation:

Formulas: Minitab, methods and formulas (6 × SD by default) and number of distinct categories. The ndc uses 1.41, the rounded value of √2 common in gauge R&R templates; Minitab uses √2 itself (Minitab, number of distinct categories). The difference only matters when the result is very close to a whole number.

Worked example: the AIAG data

The calculator opens with this example. It is the data set of the AIAG MSA manual (4th edition), published independently by Minitab and SPC for Excel, so you can check every number. Here is what they publish and what the calculator shows:

The ndc of 5 for average and range (1.41 × 1.1044 / 0.30578 = 5.09) is our own calculation; the source doesn’t give it. The last digits differ slightly from a hand calculation with the printed constants (0.5908, 0.5231, 0.3146) because the calculator uses the unrounded values.

What the example teaches:

• 7.76 % and 27.86 % are the same result on two scales. One is a share of variance, the other a share of standard deviation: 0.2786² ≈ 0.0776.
• Judged on %Study Variation, the gauge is marginal by either method, and its ndc of 4 is below the AIAG minimum of 5. SPC for Excel (2015) notes that the AIAG manual itself calls the ANOVA result acceptable because the %GRR is under 10 %, which applies the band to the variance scale. The calculator follows Minitab and applies the bands to %Study Variation and %Tolerance (Minitab, interpreting the results).
• Against the tolerance, the gauge does a little better: it uses 22.68 % of the tolerance and 27.86 % of the process variation. Which one matters depends on what the gauge is for.

Reading the verdict

The calculator labels %Study Variation and, if you entered a tolerance, %Tolerance, with the AIAG bands, in our words:

We haven’t read the AIAG manual itself, which is sold by AIAG; the criteria above come from two independent publications that quote it, and we rate them as reliable but second-hand. Check what your customer requires: many automotive customers specify their own acceptance rules.

• %Study Variation answers “can the gauge see the variation of this process?” and matters for SPC and process control.
• %Tolerance answers “can the gauge tell good parts from bad ones?” and matters for inspection against a specification.
• %Contribution is not compared with the 10 / 30 % bands: on the variance scale, 10 % corresponds to about 32 % of the standard deviation. The Minitab Blog (2011) gives 1 % and 9 % as the matching cutoffs for that column.

If the study fails, the components point at the cause: a high repeatability suggests the gauge, fixture or part form, and a high reproducibility suggests the method or training. The guide has a table of what to do when a gauge R&R fails.

Why your spreadsheet or other software gives a different answer

Different results for the same data are usually not errors. These are the settings that change them:

Sources: Minitab (example; methods and formulas); SPC for Excel (2015; ANOVA Gage R&R Part 1); SAS/QC (multiplier of 4, 5.15 or 6); Ermer (2006, 5.15σ and d2*). Older material used 5.15 × SD, which covers about 99 % of a normal distribution, instead of 6 × SD (Ermer, 2006; SAS/QC).

Assumptions and limits

• Crossed, balanced studies with numeric readings. Every operator measures every part the same number of times. Destructive tests need a nested design, and pass/fail gauges need an attribute agreement study (see the guide).
• Random model, normal errors. ANOVA treats parts and operators as random samples. The p-values assume roughly normal measurement errors; they are checked against the F table of the NIST e-Handbook (§1.3.6.7.2).
• Parts must represent the process. %Study Variation and ndc divide by the variation of the parts in the study. Parts that barely differ make a good gauge look bad, and parts far outside the normal range make any gauge look good (Minitab, methods and formulas). With fewer than 5 parts the calculator warns you.
• No historical process variation. Some software can divide by a known process standard deviation instead of the study’s part variation. This version doesn’t.
• Range constants for any size. The average and range constants for sizes beyond the published examples are computed, not taken from a printed table; they agree with the published d2, d3 and d2* tables (Ermer, 2006) to the precision printed there.
• Validation. The math is checked by 143 automated tests against the AIAG example (Minitab and SPC for Excel), a second SPC for Excel ANOVA example with the interaction kept and a negative variance set to zero, the Ermer (2006) example with 2 trials, and published d2, d3 and d2* tables. A further 37 tests check how pasted data is read.
• Calibrate first. A gauge R&R measures precision, not bias: accuracy is a matter of calibration, done before the study (Ermer, 2006). A gauge that reads consistently off goes unnoticed, because shifting every reading by the same amount leaves every variance component unchanged.