Gauge R&R Studies: How to Assess Measurement Error

Every measurement contains error, and these errors can cost scrap, rework, and missed quality targets. A gauge R&R study splits the spread in a set of readings into part-to-part spread and gauge spread (repeatability and reproducibility), so a team can see how much of the spread comes from the gauge.

Worked Example (Simulated): How Much of the Spread Is the Gauge?

Question: How much of the spread in these bore reads comes from the gauge?

Inputs: Made-up data, seed 271828. The script makes reads for five parts, three staff, and two tries, so there are thirty reads. It sets part and staff effects, then adds random gauge noise.

Method: Use a balanced ANOVA model. ANOVA is a way to split spread by source. Each staff member reads each part, so this is a crossed study. Split spread into parts, staff, staff-by-part effects, and repeat noise, and get each piece from the mean squares. Add the gauge pieces for GRR (gauge repeatability and reproducibility). Report GRR as a share of total standard deviation (SD, a measure of spread): GRR SD divided by total SD. The script also prints the share of variance. This is not a job limit.

import random
from math import sqrt

seed = 271828
rng = random.Random(seed)
parts = range(5)
operators = range(3)
tries = range(2)
part_effect = [-0.02, -0.01, 0.0, 0.01, 0.02]
operator_effect = [-0.002, 0.0, 0.002]

data = {
    (p, o, t): 25.0 + part_effect[p] + operator_effect[o] + rng.gauss(0, 0.0012)
    for p in parts for o in operators for t in tries
}
P, O, R = len(parts), len(operators), len(tries)
mean = sum(data.values()) / len(data)
pmean = [sum(data[p, o, t] for o in operators for t in tries) / (O * R)
         for p in parts]
omean = [sum(data[p, o, t] for p in parts for t in tries) / (P * R)
         for o in operators]
cmean = {(p, o): sum(data[p, o, t] for t in tries) / R
         for p in parts for o in operators}

ss_part = O * R * sum((x - mean) ** 2 for x in pmean)
ss_op = P * R * sum((x - mean) ** 2 for x in omean)
ss_cell = R * sum((cmean[p, o] - mean) ** 2
                  for p in parts for o in operators)
ss_inter = ss_cell - ss_part - ss_op
ss_error = sum((data[p, o, t] - cmean[p, o]) ** 2
               for p in parts for o in operators for t in tries)

df_part, df_op = P - 1, O - 1
df_inter, df_error = (P - 1) * (O - 1), P * O * (R - 1)
ms_part, ms_op = ss_part / df_part, ss_op / df_op
ms_inter, ms_error = ss_inter / df_inter, ss_error / df_error

v_repeat = ms_error
v_inter = max(0.0, (ms_inter - ms_error) / R)
v_operator = max(0.0, (ms_op - ms_inter) / (P * R))
v_part = max(0.0, (ms_part - ms_inter) / (O * R))
v_grr = v_repeat + v_inter + v_operator
v_total = v_grr + v_part
pct_grr = 100 * sqrt(v_grr / v_total)

print(f"seed={seed}; parts={P}; operators={O}; tries={R}; readings={len(data)}")
print("Variance components (mm^2):")
print(f"  part-to-part: {v_part:.9f}")
print(f"  repeatability: {v_repeat:.9f}")
print(f"  operator-by-part: {v_inter:.9f}")
print(f"  operator: {v_operator:.9f}")
print(f"  total: {v_total:.9f}")
print(f"GRR share of total SD: {pct_grr:.2f}%")
print(f"GRR SD: {sqrt(v_grr):.6f} mm; total SD: {sqrt(v_total):.6f} mm")
print(f"GRR share of total variance: {100 * v_grr / v_total:.2f}%")

Printed output:

seed=271828; parts=5; operators=3; tries=2; readings=30
Variance components (mm^2):
  part-to-part: 0.000250311
  repeatability: 0.000001599
  operator-by-part: 0.000000000
  operator: 0.000007054
  total: 0.000258964
GRR share of total SD: 18.28%
GRR SD: 0.002942 mm; total SD: 0.016092 mm
GRR share of total variance: 3.34%

The operator-by-part line prints 0.000000000 because its estimate came out negative and the script sets a negative estimate to zero.

Result: In this run, GRR is 18.28% of total SD. GRR SD is 0.002942 mm and total SD is 0.016092 mm. As a share of variance, GRR is 3.34%. Both figures describe the same run; the variance share is the lower one because squaring a share below one makes it smaller. A team can review the gauge with this result.

Limit: The data are made up and the model is simple. The part and staff effects were typed into the script, so the split follows them. Five parts leaves 4 degrees of freedom for the part-to-part figure, so that figure is rough. This does not show if a real gauge fits a job, and it does not predict gains.

Introduction

In manufacturing and quality control environments, decisions are made based on measurements. Part dimensions, chemical concentrations, electrical properties, and countless other parameters guide whether products ship to customers or get scrapped. But what happens when the measurement system itself is the problem?

A gauge R&R study quantifies how much of the variation in your measurement data comes from the measurement system versus the actual parts being measured. The "R&R" stands for Repeatability and Reproducibility, two fundamental sources of measurement variation. Repeatability measures how consistent a single operator is when measuring the same part multiple times. Reproducibility measures how consistent different operators are when measuring the same parts.

Measurement error can lead a team to reject good parts or pass bad ones. A gauge R&R study shows how much of the spread in readings comes from the gauge and the people using it, which can help a team decide where to look first.

What Are Gauge R&R Studies?

Gauge R&R studies use analysis of variance (ANOVA) to partition total measurement variation into components. When you measure a part, the observed value contains three types of variation:

• Part-to-part variation: The actual differences between parts
• Repeatability (Equipment Variation): Variation from the measurement instrument itself
• Reproducibility (Appraiser Variation): Variation between different operators using the same equipment

Gauge R&R analysis adds these components as variances (squared spreads):

Total variance = part variance + repeatability variance + reproducibility variance

The goal of a gauge R&R study is to determine what share of total variation comes from the measurement system (repeatability plus reproducibility). The worked example above reports that share two ways: as a share of standard deviation and as a share of variance.

A gauge R&R study typically follows a crossed design where multiple operators measure the same parts multiple times. The worked example above uses 5 parts, 3 operators, and 2 tries.

When to Use This Technique

Gauge R&R studies deliver maximum value in specific situations. Knowing when to invest time in a study versus when a simpler check will do is crucial for resource allocation.

High-Value Applications

Run a gauge R&R study when checking new measurement equipment before production use. A coordinate measuring machine (CMM) that cannot reliably distinguish good parts from bad parts wastes capital and can create ongoing quality problems. A study before you accept new equipment may also give you data to bring to the vendor.

Use gauge R&R studies when investigating persistent quality issues. If scrap rates remain high despite process improvements, measurement error may be creating false rejects.

Conduct gauge R&R studies before implementing Statistical Process Control (SPC). SPC systems monitor process variation using control charts, but these charts only work if measurement variation is small relative to process variation. Installing SPC on a measurement system with poor gauge R&R wastes resources and produces misleading signals.

Customer and Audit Requirements

Some customers and industries ask for measurement system analysis. Check the current requirements that apply to your work.

Beyond any requirement, gauge R&R studies give you a record of how a measurement system performed. When someone questions quality data, that record shows what was checked.

Process Improvement Initiatives

Many Six Sigma projects include measurement system analysis early. The DMAIC (Define-Measure-Analyze-Improve-Control) method calls for checking the measurement system before collecting baseline data. A process capability index (Cpk) calculated from unreliable measurements can mislead improvement efforts.

When implementing linear regression models or other statistical analyses, measurement error affects model accuracy and coefficient estimates. Gauge R&R studies quantify measurement error, enabling analysts to adjust models or collect additional data to compensate for measurement limitations.

Key Assumptions and Requirements

Gauge R&R studies make several statistical assumptions. Breaking these assumptions can make the results misleading.

Statistical Assumptions

The ANOVA method assumes that measurement errors follow a normal distribution. Most measurement systems satisfy this assumption approximately, but highly skewed error distributions require data transformation before analysis. Testing residuals for normality, using probability plots or statistical tests, checks this assumption.

The analysis assumes that measurement variation is stable across the measurement range. If repeatability degrades at the high end of the measurement range, a single gauge R&R percentage misrepresents system performance. In this case, conduct separate studies across different measurement ranges or use more sophisticated variance component models.

Independence of measurements is critical. Operators should not see previous measurement results when taking repeated measurements. This prevents bias where operators unconsciously try to match previous readings. Randomize measurement order and conceal previous values during data collection.

Study Design Requirements

Select parts that span the full range of expected production variation. If you select 10 nearly-identical parts, part variation will be artificially low, making measurement system variation appear larger by comparison. Choose parts representing the typical production distribution.

Use operators with normal training and experience. Do not select your best operator or your newest operator. The goal is to characterize typical measurement system performance, not best-case or worst-case scenarios.

Ensure measurement conditions match normal production. If production measurements occur at room temperature but the gauge R&R study happens in a climate-controlled lab, results will not represent actual measurement system performance. Temperature, humidity, lighting, and other environmental factors should match production conditions.

Where Gauge Error Can Cost Money

Gauge error can cost money in a few ways. Writing down which ones apply to your process helps you decide whether a study is worth running.

Scrap and Rework Reduction

Poor measurement systems create Type I errors (rejecting good parts) and Type II errors (accepting bad parts). Type I errors generate scrap costs. The larger the gauge R&R percentage, the more good parts near a specification limit are measured as out of tolerance and scrapped.

Improving gauge R&R through equipment upgrades or procedural changes reduces false rejections. Compare the cost of the improvement with the scrap you can attribute to measurement error in your own process.

Warranty and Quality Cost Reduction

Type II errors (accepting bad parts) create warranty costs, returns, and customer complaints. An automotive supplier shipping defective parts faces warranty claims, recall costs, and potential customer loss. A gauge R&R study that identifies measurement system problems preventing detection of borderline defects can prevent downstream warranty and return costs.

Process Capability Improvement

Measurement variation lowers observed process capability. The relationship between actual process capability (Cpk-actual) and observed process capability (Cpk-observed) depends on the gauge share of total standard deviation (GRR, as a fraction):

Cpk-actual = Cpk-observed / sqrt(1 - GRR^2)

The larger the gauge share, the larger the gap between observed and actual capability. A better measurement system can show process shifts sooner, which may allow faster corrective action.

What a better measurement system is worth depends on your process, and sizing it takes process-specific analysis.

Interpreting Results

Gauge R&R study output includes several metrics. Understanding each metric and its implications guides improvement decisions.

Percentage Gauge R&R (%GRR)

This page reports percentage gauge R&R as the gauge share of total standard deviation:

%GRR = 100 * (GRR SD / Total SD)

The share of variance is a different number: it is the square of the share of standard deviation. Do not mix the two. The worked example prints both. Whether a value is acceptable depends on the job, the tolerance and the cost of a wrong call, and this page gives no cutoff.

Repeatability vs. Reproducibility

The gauge R&R calculation separates repeatability (equipment variation) from reproducibility (operator variation). This separation guides improvement efforts.

High repeatability (equipment variation dominates) indicates equipment problems: worn measurement surfaces, inadequate resolution, environmental sensitivity, or inherent equipment limitations. Solutions include equipment maintenance, environmental controls, or equipment replacement.

High reproducibility (operator variation dominates) indicates training issues, inconsistent measurement procedures, or subjective judgment requirements. Solutions include improved training, better measurement procedures, fixturing to reduce operator influence, or automation.

Number of Distinct Categories (ndc)

Number of distinct categories represents how many groups the measurement system can reliably distinguish. Calculate ndc as:

ndc = 1.41 * (Part SD / GRR SD)

A larger ndc means the measurement system can sort parts into more groups. If ndc = 3, the measurement system can sort parts into three categories (small, medium, large) but cannot make finer distinctions.

Graphical Analysis

Gauge R&R software produces several diagnostic plots:

• Measurement by Operator: Shows if operators have different average readings (reproducibility issue)
• Measurement by Part: Shows if the measurement system can distinguish between parts
• Operator-Part Interaction: Shows if operator differences depend on which part is measured
• Range Charts: Show if measurement variation is consistent across operators

These plots often reveal problems not apparent in summary statistics. For example, one operator may have excellent repeatability while another has poor repeatability, indicating a training issue rather than equipment problem.

Common Pitfalls and How to Avoid Them

Gauge R&R studies can produce misleading results if conducted incorrectly. Knowing common mistakes can save wasted effort.

Inadequate Part Variation

A common error is selecting parts with insufficient variation. If parts are too similar, calculated gauge R&R will be artificially high. This occurs because gauge R&R percentage is relative to total variation, if part variation is small, measurement variation appears large by comparison.

To avoid this problem, select parts that cover the range the process actually makes. Parts across the full specification range can bring linearity issues if measurement system performance varies across the range, so check for that.

Operator Awareness

When operators know they are being evaluated, they often take extra care, producing better repeatability than normal production conditions. This makes the measurement system appear more capable than it actually is.

Minimize this effect by conducting gauge R&R studies as routine activities rather than special events. Do not announce studies in advance. Mix gauge R&R measurements with normal production measurements when possible. Consider repeating studies on a schedule rather than running one-time events.

Measurement Procedure Variation

Inconsistent measurement procedures inflate reproducibility. If operators use different clamping pressures, different measurement locations, or different measurement sequences, reproducibility suffers even if the equipment itself is capable.

Document detailed measurement procedures before conducting gauge R&R studies. Include specific instructions on part orientation, measurement location, clamping force, stabilization time, and reading interpretation. Train all operators on these procedures before beginning the study.

Environmental Variation

Temperature, humidity, vibration, and other environmental factors affect measurements. If environmental conditions change during the study, results will incorrectly attribute environmental variation to repeatability or reproducibility.

Conduct gauge R&R studies in stable environmental conditions. For dimensional measurements, allow parts and equipment to stabilize to ambient temperature before measuring. Monitor and record environmental conditions during the study. If conditions vary significantly, include environmental factors in the analysis or repeat the study under controlled conditions.

Best Practices

Running gauge R&R studies on a plan can help keep measurement systems capable over time.

Prioritize Critical Measurements

Not all measurements deserve gauge R&R studies. Focus resources on measurements that drive quality costs. Identify the few measurements that contribute most to scrap, rework, and customer complaints. Those measurements are the first candidates for a full gauge R&R study.

Create a measurement criticality matrix considering:

• Safety implications of measurement errors
• Customer-specified characteristics requiring documented measurement system analysis
• Historical scrap and rework costs associated with the characteristic
• Tolerance relative to process capability (tight tolerances require better measurement systems)

Establish Baseline Performance

Run initial gauge R&R studies on critical measurements to set baseline performance. This baseline serves multiple purposes: identifying immediate problems requiring correction, creating a reference for trending measurement system performance over time, and prioritizing equipment replacement investments.

Document baseline results in a measurement system capability matrix. Include gauge R&R percentage, repeatability, reproducibility, ndc, and study date. Update this matrix annually or whenever equipment or procedures change.

Implement Periodic Verification

Measurement system performance degrades over time due to equipment wear, operator turnover, and procedure drift. Schedule periodic gauge R&R checks. Set the schedule to fit how critical each measurement is.

Tracking gauge R&R results over time can give early warning of a gauge that is getting worse before quality problems show up.

Link Gauge R&R to Equipment Replacement Decisions

Use gauge R&R results to support equipment replacement requests. Include gauge R&R data showing the current system and the expected result with new equipment, and estimate the cost effects from your own scrap and rework records.

Train Operators on Measurement Fundamentals

Operator training directly impacts reproducibility. Develop standardized training programs covering:

• Proper measurement technique for specific equipment
• Environmental factors affecting measurements (temperature, cleanliness, stabilization)
• Reading interpretation and recording procedures
• Recognition of measurement anomalies indicating equipment problems

Incorporate gauge R&R concepts into training. When operators understand that their measurement consistency is quantified and tracked, they may improve focus and technique. Some organizations use individual operator reproducibility scores as training metrics.

Integrate with Statistical Process Control

Gauge R&R studies and SPC are complementary. Before implementing control charts, check that the measurement system is adequate for the job. Otherwise, control charts can trigger false alarms based on measurement variation rather than process changes.

Standard control chart formulas assume negligible measurement error. When gauge error is large, consider adjusting control limits to avoid false alarms. Statistical software packages often include measurement-adjusted control chart options.

Related Techniques and Extensions

Gauge R&R studies are part of a broader measurement system analysis toolkit. Understanding related techniques enhances measurement system optimization.

Attribute Agreement Analysis

Gauge R&R studies apply to continuous measurements (dimensions, weights, temperatures). For attribute data (pass/fail, defect classifications), use attribute agreement analysis. This technique quantifies operator consistency when classifying parts into categories.

Attribute agreement analysis is critical for visual inspection, defect classification, and any measurement where operators assign discrete categories. Poor attribute agreement creates inconsistent accept/reject decisions, leading to scrap variation and customer quality issues.

Gauge Linearity and Bias Studies

Standard gauge R&R studies assume measurement system performance is consistent across the measurement range. Linearity studies check this assumption by evaluating measurement accuracy at different points in the operating range. Bias studies quantify systematic measurement offset.

Conduct linearity and bias studies when measurements span wide ranges or when accuracy requirements vary across the range. These studies use reference standards with known values to quantify measurement system accuracy in addition to precision.

Nested Gauge R&R

Standard gauge R&R uses a crossed design where all operators measure all parts. In destructive testing or when parts cannot be measured multiple times, use nested gauge R&R designs where each operator measures different parts.

Nested designs require larger sample sizes to achieve equivalent statistical power and cannot estimate operator-by-part interaction. However, they enable gauge R&R analysis in situations where crossed designs are impossible.

Expanded Gauge R&R with Additional Factors

Advanced gauge R&R studies can include additional variation sources: multiple measurement systems, environmental conditions, or part characteristics. For example, a gauge R&R study evaluating two different measuring instruments simultaneously separates equipment effects from operator effects.

These expanded studies require more sophisticated experimental designs and analysis but provide comprehensive understanding of measurement system performance drivers. Use expanded studies when optimizing measurement systems or comparing alternative measurement technologies.

Conclusion

A gauge R&R study quantifies measurement system variation and separates equipment effects from operator effects. That can point to specific changes worth testing.

A sound approach starts with a baseline for critical measurements, continues with analysis and improvement, and keeps checking over time.

A measurement system with low error can support tighter process control and better decisions that depend on measured data.

To start, pick the three measurements with the highest quality cost in your operation and run a baseline gauge R&R study on each. Estimate what improvement could be worth using your own records, and bring that estimate to management with specific proposals.