Every data-driven decision assumes the data is accurate. Often it isn't — not because the process is bad, but because the measurement is. Measurement System Analysis (MSA) is the discipline of validating that your measurements can be trusted before you act on them. It checks the ruler before judging the part.
Why measurement itself is a process
A measurement is the output of a system — the gauge, the operator, the method, the environment — and like any process it has variation. If the variation in your measurement is large relative to the variation you're trying to detect in the product, your data is noise dressed as signal. MSA quantifies how much of the variation you see is real versus measurement error.
The key properties
- Repeatability — same operator, same part, same gauge: do you get the same reading? (variation within one measurer)
- Reproducibility — different operators measuring the same part: do they agree? (variation between measurers)
- Bias — does the gauge systematically read high or low versus a known reference?
- Linearity and stability — does accuracy hold across the range, and over time?
Garbage in, garbage out — but the garbage often enters at the measurement, not the process. MSA is how you catch it.
Gage R&R
The workhorse study is Gage R&R (repeatability and reproducibility), which separates total variation into part-to-part variation (what you care about) and measurement-system variation (what you want small). If the measurement system eats too much of the total, you fix the measurement before trusting any conclusions drawn from it.
Where it fits
MSA is a pillar of Six Sigma and quality engineering, and the logic generalises far beyond manufacturing: any time you make decisions from measured data — quality, lab results, KPIs, surveys — it's worth asking whether the measurement system is good enough to support the decision. Validating the data is cheaper than acting on bad data.
