How to Do a Gauge R&R Study: Measurement System Analysis Explained Step by Step

Why calibration is not enough. The case for Gauge R&R

Most quality managers understand calibration: you send the instrument to an accredited laboratory, it is compared against a traceable reference standard, and you receive a certificate confirming the instrument's accuracy. But calibration answers only one question: does this instrument read the true value correctly? It does not answer the more operationally important question: can this measurement system reliably distinguish between good parts and bad parts in the conditions under which it is actually used?

That second question is what Gauge R&R answers. An instrument can be perfectly calibrated (accurate to within its specification), but if its resolution is too coarse for your tolerance, or if two operators using the same instrument consistently get different readings, your measurement system is not capable. You will either accept bad parts (because the imprecision masks defects) or reject good parts (because the imprecision makes conforming parts look nonconforming). Either outcome is costly.

Gauge R&R is mandated in automotive supply chains under IATF 16949 and the AIAG MSA manual, but it is equally valuable (and increasingly expected), in medical device, aerospace, and precision electronics manufacturing. Even without a formal mandate, any organisation using measurements for product release decisions should periodically verify that the measurement system can do the job it is being asked to do.

Calibration vs Gauge R&R. A clear distinction

Understanding the relationship between calibration and Gauge R&R prevents a common misunderstanding: that calibration is sufficient for measurement system assurance.

Both are necessary. Calibration confirms accuracy; Gauge R&R confirms precision and capability. A measurement system needs both to be trustworthy.

The two main methods. Crossed vs Nested Gauge R&R

There are two standard Gauge R&R designs, each suited to different situations.

Crossed (ANOVA) Gauge R&R

This is the standard method. Each operator measures every part multiple times. The design is "crossed" because every operator-part combination is measured. This is the method to use when: (a) parts can be measured multiple times without being altered or destroyed, (b) you have 2–3 operators and 10 parts, and (c) you want to separate operator-to-operator variation (reproducibility) from instrument-to-instrument variation within a single operator's measurements (repeatability). The ANOVA method is preferred over the older Range method because it isolates the interaction between operator and part (some operators may be more consistent on certain part geometries), which the Range method cannot detect.

Nested Gauge R&R

This is used when parts cannot be re-measured by multiple operators because measurement is destructive or alters the part. In a nested design, each part is measured only by one operator. Common examples include: tensile strength testing (the part breaks), hardness testing (the indentation changes the surface), and chemical concentration analysis (the sample is consumed). In a nested design, reproducibility between operators cannot be separated from part-to-part variation in the same way, so interpretation requires care.

For most non-destructive measurement applications in manufacturing (dimensional, electrical, temperature, pressure), the Crossed Gauge R&R is the appropriate choice.

Step-by-step. How to conduct a Crossed Gauge R&R study

Step 1: Define the study scope

Identify the measurement system under study: the specific instrument, the measurement parameter (for example, outer diameter in mm), and the tolerance (for example, 25.00 ± 0.05 mm). The tolerance is the critical input , %GRR is calculated as a percentage of the tolerance (or total study variation, depending on method). Without a defined tolerance, the study cannot be evaluated.

Step 2: Select 10 representative parts

The 10 parts should span the full range of expected product variation, from the low end of tolerance to the high end. This is important: if all 10 parts are very similar (close to nominal), the Gauge R&R calculation will show that measurement variation is a high percentage of product variation, which may lead to an artificially poor result even if the gauge is perfectly capable. The parts should represent real production variation, not be drawn from a single batch.

Step 3: Select 2–3 operators

Choose operators who actually use the instrument in normal production or inspection. Do not select operators specifically because they are consistent. The study must represent the real measurement system as it operates. Three operators is preferred for richer data, but two is acceptable.

Step 4: Number the parts

Mark each part with a discreet identifying number. The operators should not know which part they are measuring by number during the study. The order of measurement should be randomised for each operator to prevent learning effects. However, the analyst tracking the data needs to know which reading corresponds to which part.

Step 5: Conduct the measurements

Each operator measures each part. Then, without seeing their previous readings, each operator repeats the measurements in a different random order. For a standard 2-replicate study, this gives: 10 parts × 3 operators × 2 replicates = 60 measurements. For 3 replicates, 90 measurements. Record all readings in the Gauge R&R data collection sheet.

Step 6: Analyse the data

Enter the data into statistical software (Minitab, JMP, Excel with QI Macros, or a dedicated Gauge R&R calculator), and run the ANOVA analysis. The software will calculate repeatability (EV (Equipment Variation), reproducibility (AV), Appraiser Variation), total Gauge R&R variation, part-to-part variation, and total variation.

Step 7: Evaluate the results

Apply the acceptance criteria below to determine whether the measurement system is capable for its intended use.

Understanding and applying Gauge R&R results

The primary output metric is %GRR. The percentage of tolerance consumed by measurement system variation.

The secondary metric is the Number of Distinct Categories (NDC). The number of non-overlapping categories the measurement system can distinguish within the product variation. A minimum NDC of 5 is required for the measurement system to be considered adequate for process control. An NDC below 5 means the gauge cannot reliably distinguish between different levels of product quality.

Interpreting the components

The ANOVA output separates measurement variation into repeatability and reproducibility.

• If repeatability is high relative to reproducibility: the problem is with the instrument itself. Poor precision, resolution too coarse for the tolerance, worn or damaged instrument, or instability in the measurement fixture. Calibration may be needed if the issue is drift; replacement or a higher-precision instrument may be needed if the issue is resolution.
• If reproducibility is high relative to repeatability: the problem is with operator technique. Different operators hold the probe differently, apply different force, read the display at different angles. The solution is training, standardised work instructions, or fixturing that eliminates operator influence.
• If both are high: the problem is systemic and requires a broader investigation.

What to do when Gauge R&R fails

A failing Gauge R&R (above 30%) is a finding, not a verdict of doom. It provides valuable diagnostic information. The correct response sequence is:

1. Diagnose the root cause using the repeatability/reproducibility split. Look at the ANOVA table to understand which source dominates.
2. If the issue is the instrument: check calibration status first. If the instrument is calibrated and in-tolerance but still failing Gauge R&R, the issue is likely resolution or inherent precision. A 0.1 mm resolution calliper cannot be expected to pass a Gauge R&R for a 0.05 mm tolerance. Upgrading to a higher-resolution instrument is necessary.
3. If the issue is operator technique: run a training session with standardised measurement procedure. Re-run the Gauge R&R after training. If the reproducibility is reduced, training was the root cause.
4. If the issue is fixture variation: implement a measurement fixture that holds the part in a consistent position and removes operator-to-part positioning variability.
5. If the issue is part variation within the study: check whether the 10 selected parts actually span the tolerance range. If all parts are very similar, re-run with parts that cover more of the tolerance.
6. Document the finding and corrective action in your measurement system records. A Gauge R&R study that found a problem and prompted corrective action is far more defensible at audit than one that was never run.

Gauge R&R and calibration. Working together

The relationship between calibration and Gauge R&R is complementary. The optimal maintenance approach for any critical measurement system is:

• Calibrate at the defined interval using a SAC-SINGLAS accredited laboratory to verify accuracy. The calibration certificate establishes the accuracy foundation (the systematic error component), on which the rest of the measurement system assurance programme depends.
• Conduct a Gauge R&R study at instrument qualification, after repair, after significant changes to the measurement process, or periodically (typically annually for critical gauges). The Gauge R&R study tests the complete measurement system (instrument, operators, environment), under real production conditions.
• Use the Gauge R&R results to inform calibration interval decisions. If a gauge consistently has excellent as-found results at calibration and excellent %GRR, its interval may be appropriate. If it shows drift or poor %GRR, both the calibration interval and the measurement process should be reviewed.

For instruments used in IATF 16949 automotive supply chains, MSA studies including Gauge R&R are a contractual requirement. For ISO 9001 and ISO 13485 environments, they represent best practice for measurement system assurance that goes beyond the minimum requirements of the standard.

The calibration certificate from an accredited laboratory serves a specific and essential role in the Gauge R&R workflow: it confirms that the instrument under study has no significant bias at the time the study is conducted. If the instrument had an uncorrected calibration error, the MSA results would be confounded. A systematic offset would appear as part of the natural variation measured by the study, and the root cause analysis would be misleading. Calibration comes first; Gauge R&R follows.

Unitest's stated expanded uncertainty values, printed on every accreditation certificate, also provide a calibration-level lower bound on what the measurement system can achieve. If our stated uncertainty for a micrometer is ±0.003 mm and your tolerance is 0.01 mm, the calibration uncertainty alone already consumes 30% of tolerance. Before any operator or environmental variation is considered. That is an early warning that the Gauge R&R study will be difficult to pass, and it may be more efficient to discuss instrument selection before committing to the full study setup.

This is the kind of technical guidance Unitest provides as part of the calibration conversation. Not just a certificate, but an informed view of whether the instrument is likely to be capable for its application. If you are setting up a Gauge R&R programme or reviewing existing measurement systems, contact us with your instrument type, measurement parameter, and the tolerance in question. We will advise on calibration requirements, uncertainty compatibility, and what to expect from the MSA study.

Attribute Gauge R&R for Pass/Fail Inspection Systems

Not every measurement system produces a continuous numeric value. Go/no-go gauges, visual inspection against a reference standard, and pass/fail functional test stations all generate attribute data, a simple accept or reject decision rather than a measured dimension. The Crossed and Nested designs described above, and the %GRR and NDC statistics they produce, do not apply to this kind of system, because there is no continuous variance to decompose into repeatability and reproducibility components. Instead, attribute measurement systems are assessed using an attribute agreement analysis, sometimes called an Attribute Gauge R&R, which evaluates whether operators agree with themselves on repeat inspections, whether operators agree with each other, and whether operators agree with a known, correct reference decision for each part.

The study design mirrors the variable Gauge R&R structure in spirit: select a set of parts (typically 20 to 30, deliberately including some clearly good, some clearly bad, and a meaningful number of borderline cases near the accept/reject boundary, since borderline parts are where inspection systems actually fail), have each operator inspect every part multiple times in random order without knowing the part identity, and record each accept/reject decision. The analysis then calculates three agreement statistics: within-appraiser agreement (does the same operator reach the same decision on repeat inspections of the same part), between-appraiser agreement (do different operators agree with each other), and agreement against the known reference standard (does the operator's decision match the objectively correct answer, established independently, for instance by variable measurement against a calibrated gauge with a documented pass/fail threshold). AIAG guidance generally expects at least 90% agreement across all three measures for a capable attribute inspection system, with particular attention paid to any systematic pattern in the disagreements, an operator who consistently passes borderline-bad parts is a different and more serious problem than one who occasionally disagrees randomly, because a systematic bias toward acceptance will let defects escape at a predictable, non-random rate.

Attribute Gauge R&R is common in Singapore manufacturing for functions such as visual cosmetic inspection, go/no-go thread gauging, and connector insertion verification, and it is frequently the weaker link in a facility's overall measurement system assurance programme precisely because it receives less attention than the numeric MSA studies mandated more explicitly by IATF 16949. A facility with excellent %GRR results on its dimensional gauges but no formal review of its visual and pass/fail inspection stations has only assured half of its actual measurement system risk.

Common Mistakes That Invalidate a Gauge R&R Study

A Gauge R&R study run with good intentions can still produce misleading conclusions if the design or execution contains one of several recurring errors. Selecting parts that are too similar to each other is the most common: if all ten parts cluster near the tolerance midpoint, the calculated part-to-part variation will be artificially small relative to measurement variation, producing a %GRR result that looks worse than the measurement system's true capability, sometimes triggering an unnecessary and expensive instrument replacement to solve a problem that does not actually exist. The opposite error, selecting parts specifically because they are easy to measure consistently, produces an artificially good result that will not hold up once the gauge is used on the full range of real production variation.

A second common mistake is allowing operators to see each other's readings or their own previous results during the study, which collapses genuine reproducibility variation into an artificially tight result, since an operator who can see that a colleague read 25.03 mm is naturally inclined toward the same figure regardless of what their own independent measurement would have shown. Randomising the measurement order and physically separating operators during data collection, small procedural disciplines, are what protect the statistical validity of the whole exercise. A third mistake, specific to facilities running their first Gauge R&R study, is failing to fix the part identification and data recording process before starting, leading to transcription errors between the physical part, its number, and the recorded reading; a single mislabelled part in a 60-measurement dataset can shift the ANOVA result meaningfully, and is precisely the kind of small operational error that a rushed first attempt at MSA is prone to making.