Gage R&R and MSA: why your data might be lying to you
Measurement System Analysis (MSA) is how Six Sigma validates that your data is trustworthy. If your measurement system can't reliably tell a good part from a bad one, your baseline is fiction — and every countermeasure built on it is fiction too.
What MSA answers
- Bias — is the measurement systematically off from the true value?
- Stability — is the measurement drifting over time?
- Linearity — is bias consistent across the range?
- Repeatability — if the same operator measures twice, does it agree?
- Reproducibility — if two operators measure, do they agree?
Gage R&R for continuous data
Gage R&R (Repeatability & Reproducibility) is the standard study for continuous measurements. 3 operators × 10 parts × 3 trials each. The output is a % contribution: total variation from the measurement system vs. total variation in the process. Rule of thumb:
| % Contribution | Verdict |
|---|---|
| < 10% | Measurement system is fine |
| 10–30% | Marginal — use with caution |
| > 30% | Unusable — fix before collecting baseline |
Kappa for attribute data
If your measurement is pass/fail (visual inspection, subjective grading), Gage R&R doesn't apply. Run a Kappa study: 3 operators × 30 parts × 2 trials each. Kappa > 0.75 is acceptable. Below 0.60 means your inspectors are essentially flipping coins — your scrap rate data is noise.
When to skip formal MSA
Light-touch PI projects often don't need Gage R&R. A useful substitute: have two independent people categorize the same 20 events. If they disagree more than 20% of the time, tighten the operational definitions before collecting any more data. This takes 30 minutes and catches most measurement problems.