What is measurement system analysis? It is a structured study that shows how much of the variation in your data comes from the act of measuring, not from the product or process itself. MSA tests the gauge, the people using it, the method, and the environment for accuracy, precision, and stability. The result tells you whether your numbers are good enough to base decisions on.
Every Lean Six Sigma project runs on data. If a caliper is worn, or two inspectors read the same part differently, you can chase a problem that does not exist or miss one that does. MSA catches that error before it spreads into your capability studies, control charts, and improvement decisions.
This guide covers why MSA matters, what it evaluates, how to run a study, and how to judge the results. It includes a worked calculation and examples from manufacturing and services. Lean Six Sigma Experts has taught and applied data-driven improvement since 2011, and MSA is the step that decides whether a baseline can be trusted.
Why measurement system analysis matters
MSA matters because every conclusion you draw from data inherits the errors of the system that produced it. Observed variation is the sum of process variation and measurement variation. When the second part is large, you cannot see the first part clearly.
Measurement error leads to wrong decisions
A weak measurement system fails in two directions. It lets bad parts pass and it rejects good parts. One costs you through customer complaints and warranty claims. The other costs you through scrap and rework. Neither shows up in a report labeled "measurement problem." It looks like a process problem, so teams waste weeks fixing the wrong thing.
Resolution is the classic trap. A gauge that reads in 0.1 mm steps cannot separate parts that differ by 0.02 mm. A common rule of thumb says a gauge should resolve at least one-tenth of the tolerance or the process spread. An operator can follow the inspection plan, use the assigned gauge, and still ship nonconforming parts because the gauge cannot see the difference.
A measurement system you cannot trust turns every later analysis into a guess with decimals.
Where MSA fits in DMAIC
MSA belongs in the Measure phase, before you collect baseline data. It then supports every later phase that depends on those numbers:
- Measure: confirm that data is collected accurately and consistently before you build a baseline or calculate process capability.
- Analyze and Improve: designed experiments and pilots need data precise enough to detect the effect you are testing for.
- Control: control charts and statistical process control are only as good as the readings behind them. A noisy gauge produces false alarms or hides real shifts.
Quality management standards such as ISO 9001 also require organizations to control their monitoring and measuring resources. A documented MSA is one of the cleanest ways to show an auditor that you do.
What a measurement system analysis evaluates
MSA evaluates five statistical properties of a measurement system: bias, linearity, stability, repeatability, and reproducibility. The first three describe accuracy. The last two describe precision. Most studies also check resolution, the smallest change the gauge can detect.
The five elements at a glance
Each element answers one specific question about your data.
| Element | Group | Question it answers |
|---|---|---|
| Bias | Accuracy | How far is the average reading from the reference value? |
| Linearity | Accuracy | Does bias change across the gauge’s operating range? |
| Stability | Accuracy | Does the average reading drift over time? |
| Repeatability | Precision | How much do readings vary when one person measures the same part repeatedly with the same gauge? |
| Reproducibility | Precision | How much do averages differ when different people measure the same part with the same gauge? |
Bias is simple arithmetic: the average of your readings minus the reference value. A system is stable only when no special cause shifts that bias over time.
Accuracy versus precision
A gauge can be precise and still wrong. Picture a bathroom scale that reads 152.0 lb every time for a person who weighs 150 lb. The readings are tight, but they are biased. Now picture a scale that reads 147, 153, 149, and 151. It is right on average but imprecise.
MSA separates the two because the fixes differ. A biased gauge needs calibration or adjustment. An imprecise one needs a better fixture, finer resolution, a clearer method, or more training.
Accuracy tells you whether the gauge is centered. Precision tells you whether you can believe a single reading.
Sources of measurement variation
The measurement system is more than the gauge. Five sources can add variation:
- Method: the test procedure and specification.
- People: operator skill, training, and technique.
- Tools: gauges, fixtures, software, and their calibration.
- Items measured: the parts or samples and how you sample them.
- Environment: temperature, humidity, lighting, vibration.
Calibration compares an instrument to a standard of known accuracy, traced back to a national standard. It corrects bias, but it says nothing about repeatability or operator effects. A freshly calibrated gauge can still fail an MSA.
How to run a measurement system analysis
Run an MSA by choosing parts that span the process spread, using the operators who normally measure, taking repeated readings in random order, and calculating what share of total variation the measurement system causes. For continuous data, the standard tool is a Gage Repeatability and Reproducibility (Gage R&R) study.
Steps for a variable Gage R&R study
Follow this sequence for a gauge that produces continuous readings:
- Select the gauge and confirm its resolution suits the tolerance.
- Pull at least 10 parts from regular production. They must cover the full process spread, low end to high end. Mark the exact measurement spot on each part.
- Choose three operators who normally perform this inspection.
- Number the parts and keep the numbers hidden from the operators.
- Have each operator measure every part three times, in random order. A third person records the data. That gives 10 parts × 3 operators × 3 trials = 90 readings.
- Calculate repeatability, reproducibility, combined Gage R&R, and part-to-part variation. Use the average and range method or ANOVA. ANOVA is the better choice because it also detects operator-by-part interaction, and software such as Minitab runs it for you.
- Compare the result with your acceptance criteria and act on it.
Attribute studies for pass/fail data
When the output is pass/fail instead of a number, run an attribute agreement study. A quick version uses 10 parts and 2 appraisers who each rate every part twice. More parts give a more reliable answer. The kappa statistic measures how well ratings agree beyond chance. Some guides accept kappa above 0.6, and stricter customers expect more, so confirm the requirement before you start.
How to read the results
The common guideline from the AIAG measurement systems manual compares measurement system variation with total variation:
| Gage R&R result | Verdict |
|---|---|
| Under 10% | Satisfactory |
| 10% to 30% | May be acceptable, depending on how important the application is, what the gauge costs, and what repair costs |
| Over 30% | Unacceptable, improve the system |
Then compare the two components. If repeatability dominates, the gauge is the problem. Repair, replace, recalibrate, or add a fixture. If reproducibility dominates, the operators are the problem. Standardize the method and retrain. State whether your percentage is against total variation or against tolerance, because the two answers differ.
A single Gage R&R percentage tells you whether there is a problem. The split between repeatability and reproducibility tells you where it is.
When to repeat the study
The AIAG manual does not set a fixed frequency. Customer requirements and risk decide. Repeat the study when:
- you introduce a new gauge or a different type of measurement system
- appraisers or tools change
- the gauge goes through calibration or preventive maintenance
- the part being measured changes significantly
- you finish improvement actions and need to confirm they worked
Many plants also repeat it annually alongside the calibration schedule.
Measurement system analysis examples
MSA applies anywhere a person or device turns something into a number or a rating. The clearest teaching case is a single gauge checked against a known reference, and the broader cases show how far the idea reaches beyond the machine shop.
Worked example: calipers on a reference shaft
An operator measures a reference shaft with a known diameter of 5.000 cm. Five readings come back: 5.01, 4.99, 4.97, 5.03, and 5.01 cm. Here is what the numbers say.
| Statistic | Value |
|---|---|
| Average reading | 5.002 cm |
| Bias (average minus reference) | +0.002 cm |
| Range | 0.06 cm |
| Sample standard deviation | 0.023 cm |
| Spread (6 × standard deviation) | 0.137 cm |
The caliper is accurate, since a bias of 0.002 cm is tiny. Precision is the problem. Suppose the feature has a tolerance of ±0.10 cm, a total band of 0.20 cm. The measurement spread of 0.137 cm consumes about 68% of the tolerance (0.137 ÷ 0.20). That is far above the 30% limit. This caliper would pass a calibration check and still be unfit to judge this feature.
This is a simplified one-operator check of repeatability. A full Gage R&R adds multiple parts and operators. The remedy here is likely a finer-resolution gauge, a fixture that holds the part consistently, or a tighter method.
Examples outside the machine shop
- Call center audits: if every auditor scores the same call to a different standard, associates are graded on noise. That is a reproducibility problem, and calibration sessions with shared scoring examples address it.
- Surveys: when a company measures advertising effectiveness by survey, the question order, the interviewers, and the way answers are recorded are all part of the measurement system.
- Vision cameras: an automated pass/fail camera gets an attribute study. Run it against parts of known condition and track false accepts and false rejects.
Mistakes that weaken an MSA study
Most failed studies fail on design, not on math. A study built on the wrong parts or the wrong people produces a clean-looking number that means nothing.
Design mistakes
- Narrow part selection. Parts from one tight lot understate part-to-part variation and distort the percentages. Sample across the full process spread.
- Hand-picked appraisers. Use the people who actually take the measurements, not your best technician.
- Visible part numbers. If appraisers can see numbers or earlier readings, they anchor on them. Randomize and hide.
- Too little data. Few parts or trials lower your confidence. Balance that against time, cost, and disruption.
Interpretation mistakes
- Treating calibration as MSA. Calibration addresses bias only.
- Ignoring resolution. A gauge with coarse steps can look repeatable simply because it rounds everything to the same value.
- Running the study once. Wear, new operators, and process changes alter the system. Repeat the study when those events occur.
Know your gauge before you trust your data
Measurement system analysis checks whether the numbers behind your decisions are accurate, precise, and stable. It splits measurement variation into bias, linearity, stability, repeatability, and reproducibility. It then compares that variation with your process or tolerance. Under 10% is satisfactory, over 30% needs action, and the middle depends on the risk. Run it before you build a baseline, and again whenever the system changes.
If you want help running MSA on your own gauges or building the skill in-house, our Green Belt Online Training Course runs about 28 hours and puts a strong emphasis on measuring. You can also contact Lean Six Sigma Experts for a free 30-minute assessment with no obligation.