The Same GR&R Data, Two Opposite Conclusions? —— Five Steps for MSA Interpretation and Judgment Criteria

Gauge Repeatability and Reproducibility (GR&R) analysis is complete, and the software provides a number. In the meeting room, the debate begins: the quality team says it exceeds the limit and the gauge must be replaced; the equipment team says it has been sufficient for a long time. Looking back at the report, both sides are actually citing the same data, but one uses "process variation" as the denominator, while the other uses "tolerance band." Different denominators naturally lead to opposite conclusions. This is not about who is being evasive; it is the easiest step to overlook in MSA interpretation—clarifying the criteria before discussing whether it is qualified or not.

1. Three Denominators, Three Rulers

The absolute value of GR&R (e.g., 0.012mm) itself does not indicate good or bad; it must be divided by a reference value to be meaningful. There are three common references:

Process Variation (Study Variation): Divide GR&R by the total variation of the characteristic (usually 6σ) to get the commonly referred to %GRR or %Study Variation. This measures whether the gauge can distinguish the variations in the process, from the perspective of SPC and capability analysis.

Tolerance Band: Divide GR&R by the tolerance band width to measure whether the measurement system is sufficient for determining conformity. This is from the perspective of inspection and PPAP.

Variance Contribution: Divide the variance of the measurement variation by the total variance (%Contribution) to reflect how much of the total variation is "consumed" by measurement error. This number is usually much smaller than the first two and is often used to "look good."

These three numbers come from the same data set, but because process variation and tolerance band are fundamentally different, the differences can be significant: when the process is very stable, the process variation is much smaller than the tolerance band. The same measurement system may be deemed nonconforming when calculated using process variation, but conforming when calculated using tolerance.

2. An Example to Clarify the Discrepancy

Consider the diameter of a certain shaft, with a drawing tolerance of 20.00±0.05mm, i.e., a tolerance band of 0.10mm. The process is well-controlled, with a total process variation of 6σ at 0.04mm. The GR&R analysis result: combined repeatability and reproducibility is 0.012mm.

Using process variation: 0.012 ÷ 0.04 = 30%, which is on the red line of the common threshold; using tolerance: 0.012 ÷ 0.10 = 12%, which is "acceptable."

The same ruler, the same data set, one judgment is a death sentence, the other is a pass. Who is right? The answer is: it depends on what the ruler will be used for next. If it is used to identify nonconforming products and for inspection release, the tolerance band is the meaningful reference; if it is used to monitor process drift and calculate Cpk, the process variation is the reference. Mixing the two criteria can lead to outcomes such as "spending hundreds of thousands on a new gauge, but the problem persists" or "the gauge is clearly not precise enough, but the inspection always passes."

3. Five Steps to Define Criteria and Conclusions

Step 1: Write the purpose first, then choose the denominator. Before the analysis, clearly define the purpose of the measurement system—whether it is for product judgment (compared to specifications) or process monitoring (compared to the process). Use the tolerance band for the former and the process variation for the latter; if both are needed, report both percentages, not just one.

Step 2: Check if the sample parts cover the true process variation. The denominator for process variation is estimated from the sample parts. If the sample parts come from the same batch, the same operator, and the dimensions are clustered together, the process variation will be severely underestimated, and the %Study Variation will be falsely high, leading to a wrongful conclusion. The sample parts should cover the actual range of process variation, and if necessary, select from different shifts and batches.

Step 3: Calculate the number of distinct categories (ndc), which is a hard metric independent of the denominator. ndc reflects how many categories the gauge can divide the process into, and it is typically required to be at least 5. If ndc is insufficient, it means the gauge cannot distinguish process differences—regardless of how good the percentage looks when divided by any denominator, process monitoring is not valid, and the gauge must be replaced, its resolution improved, or the measurement method changed. This is a requirement that cannot be bypassed by "choosing the denominator."

Step 4: Handle results by intervals, don't treat thresholds as regulations. The common rule is to pass if it is below 10%; 10% to 30% is decided based on the importance of the characteristic, the cost of the gauge and modifications; over 30% generally requires improvement. The 10% and 30% thresholds are empirical boundaries, not rigid regulations: for safety, regulatory, and key characteristics, stricter requirements can be set; for general dimensions with other means of backup, a more lenient acceptance can be allowed. Regardless, the reasons and risks for acceptance must be documented.

Step 5: Include the criteria, conclusions, and re-evaluation cycle in the documents. Note in the report the criteria used for this judgment, the source and quantity of the sample parts, the analysis date, and the conclusion. Include the re-evaluation cycle for the gauge in the gauge management plan. Re-evaluate after changing the gauge, inspection tool, measurement method, inspection software, or significant personnel changes.

4. Three Common Pitfalls

Reporting only one percentage without the denominator. Clients and auditors cannot judge or trace the results without knowing the denominator. Develop the habit of reporting "number + denominator" together.

Judging the gauge as nonconforming based on process variation in a very stable process. The more stable the process, the smaller the process variation, and the worse the %Study Variation looks. In such cases, ask: is the gauge used to identify nonconforming products or to track process drift?

Using the tolerance band to mask insufficient resolution. A wide tolerance band makes any gauge appear "conforming," but even a small process shift will go unnoticed. Passing the tolerance band does not mean ndc is met.

Applying continuous criteria to attribute or destructive measurements. For attribute gauges like pass/fail judgments, there is no comparable numerical variation to divide by the denominator; consistency analysis and Kappa should be used instead. For destructive tests like tensile, destructive, and burn tests, where each sample can only be used once, repeatability cannot be measured, and nested designs or short-term repeatability should be used instead. In these two scenarios, applying the 10%/30% rule results in numbers that lack explanatory power.

5. One Sentence Summary

The quality of GR&R does not lie in the percentage, but in what you use as the denominator and what you intend to measure with the ruler; clarify the criteria first, and the judgment criteria will be meaningful.

The same GR&R data can lead to two conclusions; choosing the wrong reference can lead to deviation. Clarify the purpose and denominator first, then check the ndc, and finally discuss whether it is conforming or not.

Knowledge code: 6.2.1

Version: v20260930

Author: QTank QTank is dedicated to providing systematic professional knowledge, methodologies, and practical tools for quality management practitioners, helping companies continuously improve their quality capabilities.