Measurement system analysis (MSA) is the study of how much of the variation in your results comes from the measurement process rather than from the thing being measured. A measurement system that contributes more variation than the differences you are trying to detect will hide them. No amount of careful work downstream recovers what it hid.
The system is everything that touches the result: the instrument, the reagents, the calibration, the operator, the procedure and the environment. MSA characterises that whole system, not just the analyser.
MSA means two related but distinct bodies of practice, and searching for the term returns both. Knowing which one you are in saves a good deal of confusion.
In manufacturing, MSA is the tradition of the Automotive Industry Action Group (AIAG). The central AIAG study is gauge repeatability and reproducibility, alongside bias, linearity and stability studies. Several operators measure several parts several times. The analysis splits the variation into part-to-part, operator and repeat components, then judges the system against the tolerance or the total variation.
In the clinical laboratory and in the development of in vitro diagnostic (IVD) devices, the same ground is covered by the CLSI EP series. CLSI is the Clinical and Laboratory Standards Institute. Precision is only the first of several characteristics, and the others — trueness, linearity, measuring interval, detection capability, interference — each have their own protocol. There are no parts and no operators measuring the same widget; there are patient samples, control materials and concentrations spanning a range.
The statistics overlap heavily. Both traditions partition variance into named components and both compare the result against a specification. The vocabulary overlaps only partly, and where it overlaps the words can mean different things. AIAG “reproducibility” is between operators, but CLSI “reproducibility” is between sites. This guide, and the rest of this section, cover the laboratory tradition.
A measured value is the true value plus whatever the measurement system added. MSA quantifies that addition, and splits it into the two kinds that behave differently.
Random error scatters results around a centre. Measure the same sample repeatedly and the values differ. That scatter is imprecision, and it is characterised by measuring the same material many times under conditions that are deliberately varied.
Systematic error shifts the whole centre. Every result is out by a similar amount in the same direction. That shift is bias, and it is characterised by comparing against something you have reason to trust.
The two need different studies because repeating a measurement reveals imprecision and hides bias. Run the same biased method a thousand times and the mean settles ever more precisely on the wrong answer. Trueness, precision and accuracy is the guide to why the two are kept apart, and where the words go wrong.
A laboratory MSA is not one study. It is a set of them, each answering one question about the measurement system, under its own CLSI protocol.
| Question | The study | Protocol |
|---|---|---|
| How much does the result scatter on repeat measurement? | Precision, split into repeatability and within-laboratory imprecision | EP05 to establish, EP15 to verify |
| Does the method read high or low, and by how much? | Method comparison, and bias at each decision point | EP09 |
| Is the response a straight line across the range? | Linearity, as deviation from a linear fit at each level | EP06 |
| Between which concentrations can a number be reported? | Measuring interval and reportable range | EP06 with EP17 |
| How low can the method go? | Detection capability — LoB, LoD and LoQ | EP17 |
| What else in the sample changes the answer? | Interference and recovery | EP07 |
| How much error can a single result carry in total? | Total analytical error, or measurement uncertainty | EP21; EP29 or ISO/TS 20914 for uncertainty |
Which of these you owe depends on whether you are establishing performance or verifying it. One question settles that: are you validating or verifying? The CLSI EP protocol roadmap puts the same set in the order the work usually runs.
Precision is not one number, and the reason is the study design. A nested design measures the same material several times within a run, over several runs, across many days. Each layer adds its own variation on top of the one below.
Partitioning the total into those named layers is a variance components analysis, and it is the same statistics the manufacturing tradition uses to separate operator from part. The layers matter because they answer different questions. Repeatability describes the best the system can do. Within-laboratory precision, which includes the day-to-day and calibration-to-calibration variation, describes what a patient result actually carries, and it is the figure to report for routine use.
Imprecision usually changes with concentration, which is why a precision study runs at more than one level. A precision profile — imprecision modelled as a function of concentration — is more useful than a single CV. The profile also sets the limit of quantitation, the lowest concentration whose imprecision still meets your goal.
Each study above has its own worked example, listed in its own guide. The CLSI EP protocol roadmap maps the whole series by the question each protocol answers. Every guide it links carries the example workbook for its own analysis, ready to open in the Analyse-it trial.
Treating MSA as a precision study. Precision is the first characteristic and the easiest to run, so it often becomes the whole exercise. A method can be beautifully precise and consistently wrong.
Running everything at one concentration. Imprecision, bias and linearity all vary across the measuring interval. A study at the middle says nothing about the rest of the range, including the decision points, which are rarely all in the middle.
Judging the system against the wrong thing. A measurement system is adequate relative to a purpose. Compare its variation against the allowable total error for that measurand, not against a general rule of thumb.
Confusing repeatability with within-laboratory precision. Repeatability is measured within a run and is always the smaller figure. Quoting it as the method’s precision understates what a patient result carries, sometimes by a factor of two.
Importing gauge R&R acceptance rules into the laboratory. The manufacturing tradition judges a measurement system against a part tolerance. A clinical laboratory judges it against a biological or clinical performance specification. The arithmetic looks similar and the criteria are not interchangeable.
Analyse-it performs the laboratory MSA studies on your own data, inside Excel:
Every feature from all five editions for 15 days. Measurement system analysis is in the Method Validation and Ultimate editions, from US$ 475 a year. Validated against NIST and CLSI reference datasets. Start from validation vs verification to settle how much of the above your method owes, or the measurement systems analysis reference guide for the technical detail.