Measurement System Analysis (MSA) is a thorough assessment of a factory’s measurement process – gauges, instruments, methods and operators – designed to quantify how much error lives in our measurements. In practice, every sensor or caliper can introduce variation, and unchecked measurement error can hide real problems or lead to bad decisions. For example, one manufacturing plant did extensive yield tests and found no clear trends – not because the process was flawless, but because their gauges had 2–3× more variation than the process itself. In other words, wrong measurements were masking real correlations. This shows why experts say MSA must precede any data-driven analysis – before using data in SPC, regression, DOE or other tools – or else “we will be making decisions based on incorrect data”. In short, MSA gives confidence that quality data underpins all our quality control decisions.
MSA breaks a measurement system into key characteristics. Each affects how much we can trust our data:
- Accuracy (Bias): How close the average measurement is to the true value. A gauge with good accuracy (low bias) reads near the correct value on average. For example, if a length block truly measures 50.000 mm but our caliper reads 50.050 mm on average, the bias is +0.050 mm. Bias (sometimes called accuracy) is precisely defined as the difference between the observed average and the true value. An ideal system has minimal bias.
- Precision: How consistently the system measures the same part under the same conditions. Precision is about the spread of repeated readings. It has two parts: repeatability and reproducibility.
- Repeatability – variation when the same operator measures the same part multiple times with the same instrument. If the readings barely move up and down, repeatability is good.
- Reproducibility – variation when different operators measure the same part with the same instrument. If different people get nearly the same result, reproducibility is good.
Together, repeatability and reproducibility quantify the overall precision (sometimes called Gauge R&R) of the measurement system. In practice, we often compute a Gage R&R study to split total variation into these two sources.
- Linearity: Whether bias stays consistent across the measurement range. A linear system has nearly the same bias at small and large values. For example, a scale might be accurate near 10 kg but read 0.5 kg high at 100 kg – that’s non-linearity. In an MSA linearity study, we measure several known reference points (e.g. calibration weights) to see if the error (bias) changes. Ideally, bias is flat across the range.
- Stability: Whether the system’s bias and precision stay constant over time. Even a perfectly linear, repeatable gauge can drift as it wears or as temperature shifts. Stability is assessed by measuring a master sample repeatedly over days or weeks. A stable measurement system shows no systematic trend in its readings.
In addition, a capable measurement system needs adequate resolution (smallest detectable increment) – typically 1/10 of the process spread – and proper calibration, but the above components are most commonly analyzed in MSA. Improving data quality means improving these aspects of accuracy and precision.
Click HERE for Lean Six Sigma & Process Improvement Training Courses
Types of MSA Studies
Manufacturers use different MSA studies depending on the situation:
- Gage R&R (Repeatability & Reproducibility) Study: This is the most common MSA. A crossed Gage R&R typically involves 2–3 operators (“appraisers”) each measuring a set of 5–10 sample parts multiple times. Statistical tools (often ANOVA) then split the measurement variation into part-to-part (the real process variation) and gauge/ operator sources. For example, a Minitab example showed 92.24% of variation came from actual part differences versus 7.76% from the measurement system. If the total Gauge R&R (sum of repeatability and reproducibility) is small – industry guidelines say under ~10% of total variation is satisfactory (10–30% may be tolerable) – the gauge is usually acceptable. Otherwise, the study tells us whether to fix the gauge (if repeatability is high) or retrain operators (if reproducibility is high).
- Bias (Calibration) Study: To check systematic error, we measure one or more known standards or “master samples.” For example, measure a 100.000 mm block 10 times. The bias is simply the average measured value minus 100.000 mm. A non-zero bias indicates the gauge is mis-calibrated. This study quickly tells us if a scale or caliper consistently reads high or low.
- Linearity Study: A linearity study combines bias checks at multiple points in the range. For instance, if measuring length on a 0–200 mm scale, we might check the bias at 0 mm, 100 mm, and 200 mm. If the bias changes significantly (e.g. +0.01 mm at low end but +0.20 mm at high end), the system is non-linear. This helps decide if a single calibration or a more complex adjustment is needed.
- Stability Study: This is a time study. Pick a reference part (or a stable process output) and measure it regularly over weeks. Plot the results on an X̅/R chart. If the measurements stay within control limits, the system is stable; if there’s a trend or shift, we have a stability issue (drift).
- Attribute Gage Study: When measurements are pass/fail (go/no-go), we use attribute agreement analysis. For example, two inspectors check 10 parts as “good/bad” twice and we compute % agreement or Cohen’s kappa. Quality-One notes that a kappa >0.6 is often required for an acceptable attribute gauge.
In practice, any new or repaired measurement system should undergo MSA (and often annually thereafter) to verify it meets requirements.
Practical Examples
Consider a metal stamping line checking hole diameter with a caliper. If every operator measures a hole 0.1 mm larger than its true size, that’s a bias issue – the caliper is mis-calibrated. If the same operator uses the caliper and sometimes gets 2.00 mm, 2.01 mm, 1.99 mm on the same part, that spread shows poor repeatability. If different operators using the same caliper get noticeably different averages, that’s a reproducibility problem. If heavier parts compress the scale differently than light parts, bias may change with part size (non-linearity).
In another example, an automotive plant used a pressure gauge that slowly drifted over weeks. Only after a stability study did engineers spot the drift and recalibrate it. Before that, control charts showed spurious trends caused by the gauge error, not the process.These examples illustrate that poor measurements often cause poor quality. In one case study, engineers adjusted processes based on bad measurements, only to make things worse. A proper MSA would have revealed that and saved scrap.
Click HERE for Lean Six Sigma & Process Improvement Training Courses
Tools and Standards
Measurement analysts rely on both guidance documents and software:
- Standards & Guidelines: The AIAG MSA Manual (4th Edition) is widely used in automotive and manufacturing. It lays out procedures, terminology, and acceptance criteria. For example, AIAG suggests having at least 5 distinct categories (resolution levels) in a Gage R&R to ensure adequate discrimination. A common rule of thumb (from AIAG and other sources) is that total Gauge R&R error under ~10% of process variation is excellent, under 30% may be tolerable depending on context, and over 30% is unacceptable. Industry quality standards like ISO 9001 also emphasize using calibrated and checked measurement systems, even if they don’t prescribe specific MSA methods.
- Software Tools: Statistical packages like Minitab, JMP, and Excel with MSA add-ins (e.g. QI Macros) have built-in Gage R&R modules. These automatically compute repeatability/reproducibility percentages and produce charts. For instance, a Minitab example found Part-To-Part variation of 92.24% and Total Gage R&R of 7.76%, indicating most variation was true process spread. Minitab also warns if the number of distinct categories is low (e.g. only 4 when 5 is recommended). Many engineers also use Analysis of Variance (ANOVA) to break down MSA variation.
- Gauges and Calibration: Of course, the physical tools matter. Calipers, micrometers, dial indicators, digital sensors, and even human judgment must be checked. Regular calibration against certified standards, good operator training, and clear measurement procedures are as important as the statistical study.
Role of MSA in Quality Control
MSA plays a foundational role in quality systems. It tells us how much trust we can put in our data. If an MSA shows high measurement error, any control chart or capability study might be invalid. As MoreSteam notes, “MSA is a critical first step that should precede any data-based decision-making”. In Six Sigma terms, the mantra is “the more error in measurement, the more error in decisions”.
When MSA is done correctly, it improves product quality and reduces waste. For example, Quality-One describes a case where an inadequate gauge resolution let bad parts slip through (“non-conforming materials”) and increased scrap. An MSA could have caught that by showing the gauge could not distinguish small but critical differences. With good MSA, organizations catch bias, fix it, retrain staff, or replace tools – preventing costly defects and rework.
In summary, Measurement Systems Analysis ensures that good data drives quality. By quantifying accuracy, precision, bias, linearity and stability, MSA gives engineers and managers confidence that control charts, capability studies, and decisions are based on reality, not measurement noise. It’s a practical, systematic tool – backed by AIAG guidelines and supported in software like Minitab – that bridges the gap between raw measurements and real product quality.
Click HERE for Lean Six Sigma & Process Improvement Training Courses