Measurement System Analysis (MSA) and Gauge R&R: How to Validate Your Measurement System Before Trusting the Data

Why measurement error matters more than most teams assume

Every measurement contains two components: the true value of the characteristic being measured, and the measurement error introduced by the measurement system. Measurement error comes from multiple sources: the gauge itself (calibration error, resolution, wear, temperature sensitivity), the operator (how they hold the part, how they read the display, how consistently they locate the part in the fixture), the part (surface finish, fixturing variation, the part’s own deformation under measurement force), and the environment (temperature, vibration, air currents).

When measurement error is large relative to the true part-to-part variation, the data stream from the measurement system is dominated by noise rather than signal. A process that appears to be varying widely may in fact be quite consistent; the apparent variation is produced by the measurement system, not the process. Conversely, a measurement system with poor resolution may mask real process shifts by rounding all readings to the same value.

The consequences of an unvalidated measurement system compound downstream. A control chart built on noisy measurement data will show false signals (triggering unnecessary process adjustments) or miss real signals (allowing defects to pass undetected). A capability study will produce artificially low Cpk values, suggesting the process needs improvement when it does not. A DMAIC project will identify the wrong root causes because the data used to locate the problem is inaccurate.

The components of a measurement system

A measurement system is not just the gauge. It includes every factor that influences the final measured value:

• The measuring instrument: Its calibration status, resolution (the smallest increment it can display), linearity (whether its accuracy is consistent across the full measurement range), and stability (whether its accuracy changes over time).
• The operator: How the operator handles, orients, and loads the part; how they read analogue displays; how consistently they apply the measurement procedure.
• The part: How much the part’s measurable characteristics vary with temperature, deform under measurement force, or differ in surface finish across the measurement location.
• The procedure: Whether the measurement procedure specifies part orientation, measurement location, clamping force, and environmental conditions clearly enough that different operators in different locations obtain the same result.
• The environment: Temperature, humidity, vibration, and light levels that affect the gauge reading or the part’s dimensions.

Gauge R&R: repeatability and reproducibility

A Gauge R&R study separates measurement system variation into two components:

Repeatability (also called Equipment Variation, or EV) is the variation in measurements obtained when the same operator measures the same part multiple times with the same gauge, under the same conditions. Repeatability error is inherent to the gauge: it represents the gauge’s own inconsistency from one measurement to the next, regardless of who is using it. High repeatability error typically indicates a gauge that needs calibration, has worn contacts, has excessive backlash, or has inadequate resolution.

Reproducibility (also called Appraiser Variation, or AV) is the variation in measurements obtained when different operators measure the same part with the same gauge. Reproducibility error reflects differences in how operators use the gauge: how they grip the part, where they position it, how they read the scale. High reproducibility error typically indicates that the measurement procedure is insufficiently defined, that operators need training, or that the gauge is difficult to use consistently.

Total Gauge R&R variation (GRR) combines repeatability and reproducibility: GRR = sqrt(EV² + AV²). This is the total measurement system variation, used to assess whether the measurement system is adequate for its intended purpose.

The crossed Gauge R&R study design

The standard Gauge R&R study (also called a crossed design because every combination of operator and part is measured) uses the following structure: select 10 parts that represent the full range of process variation; select 2 to 3 operators who regularly use the gauge; have each operator measure each part 2 to 3 times in random order, without seeing the other operators’ results or their own previous results on the same part. This produces a dataset of (parts x operators x replications) measurements, typically 20 to 90 data points, from which the ANOVA (Analysis of Variance) method separates the variance contributions of parts, operators, the operator-by-part interaction, and gauge repeatability.

The parts selected for the study must represent the actual process variation, not just the convenient or average parts. If parts are selected from a narrow range (all near the nominal value), the study will artificially inflate the %GRR because the denominator (part variation) is small. The study should capture the range that the gauge will actually encounter in production use.

Interpreting Gauge R&R results

Gauge R&R results are most commonly expressed as a percentage of two reference values:

%GRR as a percentage of total variation (also called %Study Variation): This compares the measurement system variation to the total observed variation in the study (parts + measurement system). The automotive industry standard (AIAG MSA manual) uses the following criteria:

• %GRR below 10 percent: acceptable. The measurement system is capable.
• %GRR between 10 and 30 percent: marginal. May be acceptable depending on the application and the cost of improvement. Action is recommended.
• %GRR above 30 percent: unacceptable. The measurement system must be improved before the data can be used for quality decisions.

%GRR as a percentage of the specification tolerance (also called %Tolerance): This compares the measurement system variation to the engineering specification range (upper specification limit minus lower specification limit). A measurement system should typically consume less than 10 percent of the tolerance range to reliably accept conforming parts and reject nonconforming ones. Consuming 30 percent or more of the tolerance makes the measurement system unreliable for accept/reject decisions at the specification limits.

Number of distinct categories (ndc): The ndc indicates how many distinct groups the measurement system can reliably distinguish within the process variation. An ndc of 1 means the measurement system cannot distinguish any parts from each other; an ndc of 5 or more is generally considered adequate for process control. The ndc is calculated as: 1.41 x (part variation / measurement system variation), rounded down to the nearest integer.

Linearity and stability studies

Gauge R&R quantifies the random variation in the measurement system. Two additional MSA studies address systematic errors:

Linearity assesses whether the gauge’s bias (the difference between its average reading and the true value) is consistent across the full range of values it measures. A gauge that reads 0.02 mm too high at the low end of its range but 0.08 mm too high at the high end has a linearity problem; its error is not constant across the measurement range. Linearity is evaluated by measuring multiple reference standards of known value that span the gauge’s operating range and plotting bias versus reference value. If the bias varies significantly with the reference value, the gauge needs recalibration or replacement.

Stability assesses whether the gauge’s bias drifts over time. A stability study takes repeated measurements of a reference standard (a master part) at regular intervals over a period of weeks or months and checks whether the average reading shifts. Drift indicates that the gauge needs more frequent calibration or that environmental factors (temperature, humidity) are affecting its performance.

Attribute Gauge R&R

When the measurement system produces a binary outcome (pass/fail, conforming/nonconforming) rather than a continuous measurement, a variable Gauge R&R study cannot be applied. Instead, an attribute agreement analysis (also called attribute Gauge R&R) is used. Operators inspect a set of parts — typically including known-good parts, known-bad parts, and borderline parts — and classify each as pass or fail. The study compares each operator’s results to a reference classification (the known correct answer) and to each other. Key metrics are: within-appraiser agreement (does the operator give the same answer when shown the same part twice?), between-appraiser agreement (do different operators agree?), and agreement with the reference standard.

Attribute gauge R&R studies are particularly important for visual inspection processes, where operator-to-operator variation is often large and poorly controlled. Low agreement rates between operators typically indicate that the inspection criteria are not clearly defined, that parts at the boundary of acceptability are ambiguous, or that operators need training on specific defect types.

Common errors in MSA

• Selecting parts that are too similar. If the 10 parts in a Gauge R&R study are all near the nominal dimension, the part variation component will be small, artificially inflating the %GRR. The study parts must represent the actual range of production variation to produce meaningful results.
• Not randomizing the measurement order. If operators measure all replications of part 1 before moving to part 2, they may remember (consciously or unconsciously) their previous readings and repeat them, underestimating the true repeatability error. Randomizing the order of measurement within the study prevents this bias.
• Using the same gauge for the MSA study that is used in production, without checking its calibration first. An out-of-calibration gauge will produce inflated bias and stability errors. The gauge should be calibrated before the MSA study to separate calibration problems from inherent gauge variation.
• Confusing the MSA result with a calibration result. Calibration verifies that the gauge reads correctly against a known standard (addressing bias). MSA evaluates the total variation of the measurement system in use (addressing precision). A gauge can be well-calibrated (low bias) but still have high R&R error (low precision). Both calibration and MSA are necessary; neither substitutes for the other.

Frequently asked questions

What is Measurement System Analysis (MSA) and why is it important?

Measurement System Analysis (MSA) is a set of statistical methods for evaluating the capability of a measurement system to accurately and consistently measure the characteristic it is intended to measure. It is important because every measurement contains both the true value of the characteristic and the error introduced by the measurement system (the gauge, the operator, the procedure, and the environment). If measurement error is large relative to the variation in the parts being measured, the data produced by the system cannot reliably support process control charts, process capability studies, or root cause analysis. Before using measurement data to make quality decisions — whether for acceptance inspection, a control chart, or a DMAIC project — the measurement system should be validated through an MSA study to confirm that it is capable of detecting the level of variation being studied.

What is a Gauge R&R study and how is it conducted?

A Gauge Repeatability and Reproducibility (Gauge R&R) study is the most common MSA study for continuous measurement data. It quantifies how much of the total observed variation in measurements comes from the measurement system (repeatability and reproducibility) versus the actual variation in the parts being measured. The standard crossed study design selects 10 parts representing the full process variation range, 2 to 3 operators who regularly use the gauge, and 2 to 3 replications per operator per part. Each operator measures each part multiple times in randomized order, without seeing other operators’ results or their own previous readings. The resulting dataset is analyzed using ANOVA to separate the variance contributions of parts, operators, the operator-by-part interaction, and gauge repeatability. The total measurement system variation (GRR) is expressed as a percentage of total study variation and as a percentage of the specification tolerance. Below 10 percent is acceptable; 10 to 30 percent is marginal; above 30 percent requires improvement.

What is the difference between repeatability and reproducibility in Gauge R&R?

Repeatability (Equipment Variation) is the variation in measurements when the same operator measures the same part multiple times with the same gauge under the same conditions. It reflects the gauge’s own inconsistency and is independent of who uses it. High repeatability error indicates a gauge problem: worn contacts, insufficient resolution, backlash, temperature sensitivity, or the need for calibration. Reproducibility (Appraiser Variation) is the variation in measurements when different operators measure the same part with the same gauge. It reflects differences in how operators use the gauge: grip, part positioning, reading technique. High reproducibility error indicates that the measurement procedure is not specific enough, operators need training, or the gauge is ergonomically difficult to use consistently. Distinguishing between the two is important for identifying the correct corrective action: high repeatability calls for a gauge improvement; high reproducibility calls for an operator training or procedure improvement.

What is %GRR and what acceptance criteria should be used?

%GRR is the total Gauge R&R variation expressed as a percentage of a reference value. It is calculated in two ways. %Study Variation (also called %Total Variation) compares the measurement system variation to the total observed variation in the study. The AIAG MSA manual acceptance criteria are: below 10 percent is acceptable; 10 to 30 percent is marginal (may be acceptable depending on the application); above 30 percent is unacceptable. %Tolerance compares the measurement system variation to the specification range (USL minus LSL). A measurement system should consume less than 10 percent of the specification tolerance to reliably classify parts near the specification limits. The appropriate criterion depends on how the gauge is used: for process monitoring (control charts), %Study Variation is the relevant metric; for accept/reject inspection at specification limits, %Tolerance is the relevant metric. Always evaluate both and apply the one that matches the gauge’s intended use.

What is the number of distinct categories (ndc) in a Gauge R&R study?

The number of distinct categories (ndc) is a metric from the Gauge R&R study that indicates how many distinct groups the measurement system can reliably differentiate within the process variation. It is calculated as 1.41 multiplied by the ratio of part variation to measurement system variation, rounded down to the nearest integer. An ndc of 1 means the measurement system cannot distinguish any parts from each other and is useless for process monitoring or discrimination. An ndc of 2 means the measurement system can only sort parts into two categories (high and low), which is insufficient for most quality applications. An ndc of 5 or more is generally considered adequate for statistical process control and process capability analysis. The ndc is essentially a measure of measurement resolution relative to the process: a gauge with low R&R variation relative to part-to-part variation will have a high ndc and can detect small real differences between parts.

What is an attribute Gauge R&R and when is it used?

An attribute Gauge R&R (also called attribute agreement analysis) is an MSA method for binary measurement systems that classify parts as pass/fail, conforming/nonconforming, or present/absent, rather than producing a continuous numerical measurement. It is used for visual inspection processes, go/no-go gauges, and any other measurement system that produces a categorical result. The study presents a set of parts (typically 30 to 50, including a range of clearly conforming, clearly nonconforming, and borderline parts) to each of two or three operators on two separate occasions, in randomized order. The study reports within-operator agreement (does the operator agree with themselves?), between-operator agreement (do operators agree with each other?), and agreement with a reference standard (known correct classification). Low agreement rates indicate that the inspection criteria are ambiguous, borderline parts are not well-defined, or operators need training on specific defect types and acceptance criteria.