Compliance and audits
Gage R&R and measurement systems analysis (MSA), explained
Gage R&R measures how much of your recorded variation comes from the measurement system itself, split into repeatability (same operator, same gage) and reproducibility (between operators). It is the core study of measurement systems analysis, usually shortened to MSA, which is how you separate the two and decide whether you can trust the numbers. This guide explains the families of MSA studies in plain terms, how Gage R&R differs from bias and linearity, what the %GRR and NDC thresholds mean, when to run which study, and how the whole thing relates to calibration.
What measurement systems analysis is
The measurement system is not just the gage. It is the gage, the person using it, the method, the fixture, and the environment, all working together to produce a reading. Measurement systems analysis is a set of studies that tell you how much of the variation you see comes from the parts themselves, and how much is noise added by that system.
This matters because you make decisions from measurements. You accept or reject parts, adjust a process, or judge whether an instrument still reads true. If a large share of your measured variation is coming from the measurement system rather than the parts, those decisions rest on noise. MSA gives you a defensible, numeric answer to a simple question: can you trust the numbers this system produces?
MSA is most strongly associated with the automotive quality world, where the AIAG Measurement Systems Analysis reference manual defines the standard methods and IATF 16949 control plans routinely call for it. But the ideas apply anywhere you measure to make a decision.
The three families of MSA studies
People often say MSA when they mean Gage R&R, but Gage R&R is only one part of the picture. MSA is a family of studies, and each one answers a different question. Knowing which is which is the difference between running the right study and wasting a morning on the wrong one. The three families below cover almost everything you will need.
- Gage R&R (repeatability and reproducibility) measures precision. It tells you how much variation the measurement system adds when different people measure the same parts more than once. It says nothing about whether the readings are correct, only whether they are consistent.
- Bias and linearity, together with stability, measure accuracy. They compare the system's readings to a known reference value. Bias is the offset at one point, linearity is how that offset changes across the range, and stability is how it drifts over time.
- Attribute agreement analysis covers judgements that are not numbers: pass or fail, go or no-go, accept or reject, or a visual grade. It measures whether appraisers agree with each other, with themselves, and with the correct answer.
Gage R&R: repeatability and reproducibility explained
Gage R&R is the workhorse of MSA. The two Rs are repeatability and reproducibility, and they capture two different sources of measurement variation.
Repeatability is the variation you get when one person measures the same part, with the same gage, several times. It is sometimes called equipment variation, because it reflects the gage and the measurement itself. If a caliper gives you three different readings on the same block, that spread is repeatability.
Reproducibility is the variation you get when different people measure the same parts. It is sometimes called appraiser variation, because it reflects differences in how people use the gage, seat the part, or read the scale. If two inspectors consistently disagree on the same parts, that gap is reproducibility.
A typical study combines both into a single measure of measurement system variation, usually written as GRR. The study design below is the common one, and keeping to it is what makes the result trustworthy.
- Choose about 10 parts that represent the real range of process variation, not 10 good parts.
- Use 2 to 3 appraisers who normally do the measurement.
- Have each appraiser measure each part 2 to 3 times, in random order, blind to their earlier readings.
- Keep the gage, method, and conditions the same as real production use.
Reading Gage R&R results: %GRR and NDC
The raw GRR number is not meaningful on its own. You compare it to something: either the total variation you observed in the study, or your tolerance. Expressed as a percentage, that is %GRR. Which base you use matters. Percent of study variation tells you how the system performs against your actual process spread. Percent of tolerance tells you how it performs against the spec you have to hold. They can give different answers, so always state which one you used.
The widely used AIAG guidelines for %GRR are easy to remember, but treat them as guidelines, not law. The right threshold depends on what the measurement is for and what a wrong decision costs.
- Under 10 percent: the measurement system is generally considered acceptable.
- 10 to 30 percent: may be acceptable depending on the importance of the application, the cost of the gage, and the cost of getting the decision wrong.
- Over 30 percent: the system is generally considered unacceptable and needs work before you rely on it.
- NDC (number of distinct categories): how many distinct groups the system can reliably tell apart across your process range. A value of 5 or more is the usual target. Below that, the system struggles to resolve real differences between parts.
Bias and linearity: the accuracy side of MSA
Gage R&R tells you whether a system is consistent. It cannot tell you whether it is correct. A gage can repeat the same wrong number all day. That is what the accuracy studies are for, and they all compare your readings to a known reference value, usually a master or reference standard with a calibrated, traceable value.
Bias is the difference between the average of your measurements and the reference value at one point. It is a systematic offset. If a reference block is known to be 25.000 mm and your gage averages 25.014 mm, the bias is 0.014 mm. Bias is the same kind of error calibration deals with, which is why the two are closely linked.
Linearity asks whether that bias stays the same across the measurement range or changes. A gage might read almost perfectly near the low end and drift high near the top of its range. You measure linearity by checking bias at several reference values across the range and seeing whether it trends.
Stability, sometimes grouped with these, is how bias behaves over time. You measure the same reference repeatedly over days or weeks and watch for drift, often on a control chart. Stability is closely related to how you set calibration intervals.
Attribute agreement analysis
Not every measurement is a number. A lot of real inspection is a judgement: pass or fail, go or no-go, accept or reject, or a visual grade like scratch or no scratch. You cannot run a numeric Gage R&R on a decision, so these judgements need their own study, usually called attribute agreement analysis or attribute Gage R&R.
The study works by having appraisers judge a set of parts whose correct answer is already known, more than once, without seeing their earlier calls. From that you learn three things: whether each appraiser agrees with themselves, whether appraisers agree with each other, and, most important, whether they agree with the known correct answer.
Agreement is often summarised with a kappa statistic, which measures agreement beyond what you would expect from chance alone. Higher is better. You also look at practical error rates. A missed bad part is usually the costliest error, so watch the miss rate closely.
- Within-appraiser agreement: does each person make the same call on the same part every time?
- Between-appraiser agreement: do different people make the same call on the same part?
- Agreement with the standard: do the calls match the known correct answer?
- Miss rate and false alarm rate: how often bad parts pass, and how often good parts are wrongly rejected.
When to run which study
You do not run every study every time. Match the study to the question in front of you and to what your quality plan or customer requires. In automotive work, MSA is typically part of new part approval and is called out on the control plan, but the same logic applies anywhere you measure to decide.
- New gage or a new measurement process going into service: run Gage R&R, and check bias and linearity against a reference.
- The measurement is a pass or fail or visual judgement: run attribute agreement analysis instead of a numeric Gage R&R.
- After a gage is repaired, moved, or its method changes: re-run Gage R&R to confirm nothing shifted.
- You suspect one operator's results differ from another's: the reproducibility part of a Gage R&R will show it.
- Readings seem consistent but possibly off: that is a bias and linearity question, not a Gage R&R one.
- On a schedule set by your control plan or customer: run whatever the plan specifies, on the stated interval.
How MSA relates to calibration, and where Axiospec fits
Calibration and MSA are often confused, because both are about trusting measurements, but they answer different questions and you need both. Calibration compares one instrument to a traceable reference standard to establish, and if needed correct, its accuracy. It is the gage against a known truth. MSA evaluates the whole measurement system as it is actually used, including the people, the method, and the parts, and it is mostly about variation in real conditions.
The two overlap at accuracy. The bias and linearity side of MSA measures the same kind of error that calibration corrects, which is why a valid MSA depends on a properly calibrated reference in the first place. Calibration is a prerequisite, not a replacement. A calibrated gage can still fail Gage R&R because of operator or method variation, and a gage with a great Gage R&R can still read a consistent but wrong offset if it was never calibrated. Calibration decision rules such as the test uncertainty ratio and guard banding sit alongside all of this, governing how you accept or reject the instrument itself.
Axiospec is a calibration management tool, and it includes an AIAG Gage R&R study and per-instrument uncertainty budgets, so the measurement analysis lives next to the calibration record for the same asset rather than in a separate spreadsheet. When you run a study, the data stays attached to the instrument, alongside its as-found and as-left readings, its certificates, and its due dates. Axiospec organizes and surfaces this evidence and does the AIAG math, but it does not decide conformance. Whether a system is acceptable, and whether your program meets a standard, is a judgement for your quality team, customer, or assessor.
The free tools below run the core Gage R&R and bias and linearity math in your browser with nothing to install. If you want the studies stored on the asset, Axiospec's free plan is the full platform capped at 50 active assets, with unlimited users, and there is a 14-day free trial with no credit card.
Put it into practice
Import your asset registry in an afternoon, log every calibration to a tamper-evident audit trail, and produce records on demand. Prefer to see it first? Take a self-guided tour with sample data, no signup required.
Axiospec is a documentation and workflow tool. It helps you keep clean, traceable, audit-ready records; certification depends on your own processes, scope, and assessor.
More like this, by email
A short series of practical calibration guides like this one. No spam, unsubscribe with one click.