Accuracy and precision are the two most confused words in measurement, and mixing them up quietly costs operators money and trust. Accuracy is how close a reading is to the true value - its trueness. Precision is how close repeated readings are to each other - the absence of scatter. They are independent: an instrument can be precise and wrong, giving a beautifully steady reading that is off by a fixed amount, and that combination is far more dangerous than obvious noise because it looks trustworthy. This page untangles the two terms, walks through a SCADA transmitter that is precise but biased, and shows how the historian trend reveals precision while only a calibration check reveals accuracy.
Accuracy vs Precision in one line: Accuracy is how close a measurement is to the true or actual value, sometimes called trueness. Precision is how close repeated measurements are to one another, regardless of whether they are correct - it describes scatter, or the lack of it. The two are independent: a reading can be precise but inaccurate (tightly repeatable yet biased) or accurate on average but imprecise (correct in the mean but scattered), which is why they must be assessed separately.
The cleanest way to hold the two apart is the target analogy. Imagine shooting at a bullseye. Accuracy is whether your shots land near the center - close to the true value. Precision is whether your shots land near each other - tight grouping - wherever that group happens to be. A tight group in the center is accurate and precise. A tight group up in one corner is precise but inaccurate: consistent, repeatable, and consistently wrong. A wide scatter centered on the bullseye is accurate on average but imprecise.
The word for the corner-group problem is bias. Bias is a systematic offset that shifts every reading the same way - a transmitter that reads two units high across the board, a scale that reads heavy, a sensor with a zero shift. Because bias affects every reading equally, it does nothing to the scatter, so a biased instrument can be extremely precise. Precision cannot see bias, and that is the crux of why the two concepts must be measured differently: repeating a measurement tells you about scatter, but repeating a biased measurement a thousand times just gives you a thousand equally wrong answers.
This is why modern metrology language often prefers trueness for closeness to the true value and reserves accuracy for the combination of trueness and precision together. For everyday work the important thing is simply to remember that they answer different questions: precision asks whether the instrument agrees with itself, and accuracy asks whether it agrees with reality. An instrument must satisfy both to be trusted, and passing one says nothing about the other.
Consider a pressure transmitter on a vessel whose true pressure is holding steady at 100 units. The transmitter has drifted and now reads a consistent 105 - a five-unit high bias. On the SCADA screen the operator sees a rock-steady 105, calm and confident. Nothing about that number invites suspicion. It does not jump, it does not jitter, it holds like a good measurement should. The transmitter is highly precise, and precision is exactly the quality that makes a reading look believable, so the bias sails straight past the operator's judgment.
This is more dangerous than an obviously noisy signal precisely because it hides. A jittery reading tells the operator something is wrong and prompts a check; a steady, biased reading tells the operator everything is fine while the process is actually five units lower than displayed. Control loops inherit the lie too - a controller holding this transmitter at a setpoint of 100 will drive the real pressure to 95, and every calculation, alarm limit, and custody figure built on the reading is off by the same hidden amount. The very steadiness that reassures the operator is what makes the error invisible.
The lesson is that you cannot judge accuracy by looking at the live value or its trend, no matter how clean it looks. Steadiness proves precision and nothing more. The only way to catch the bias is to compare the transmitter against a known reference - a calibrator, a deadweight tester, a trusted standard - and that comparison is a fundamentally different activity from watching the screen. An instrument that has never been checked against a reference has never had its accuracy tested, however long it has been giving satisfyingly stable numbers.
A SCADA historian is superb at revealing precision and blind to accuracy, and knowing which is which changes how you read it. When a cloud SCADA platform such as Merobix trends a transmitter over hours or days, the amount of scatter and drift in that trace is a direct picture of the instrument's precision and stability. A trace that holds tight is precise; one that jitters or wanders is not. You can diagnose a noisy sensor, a loose connection, or a sensor whose readings are creeping, all from the trend, without ever leaving the control room.
What the trend cannot show is whether that tidy line is at the right value. The historian faithfully records whatever the transmitter sends, bias and all, so a perfectly biased instrument produces a perfectly clean, perfectly wrong trend. No amount of trend analysis distinguishes a true 100 from a biased 105 that reads steady - both look identical on screen. Accuracy lives outside the data path; it can only be established by injecting a known input and seeing what the instrument reports, which is what a calibration check does.
The practical division of labor is therefore straightforward. Use the historian to monitor precision and stability continuously and to flag when a reading starts drifting or scattering, which is often the first sign that a calibration is slipping. Use scheduled calibration checks against a traceable reference to establish and restore accuracy, on an interval appropriate to how much the process depends on that measurement. The historian tells you when something changed; the calibration tells you whether the number was ever true. For custody, safety, and control-critical points, both disciplines matter, and confusing one for the other is how a precise-but-wrong transmitter goes uncorrected for months.
Yes, and it is a common and dangerous case. An instrument with a fixed bias - reading a constant amount high or low - repeats that same wrong value consistently, so it is highly precise while being inaccurate. Because bias shifts every reading equally, it does not add scatter, which is exactly why a precise but biased reading looks trustworthy on a screen and slips past operators.
No. A trend reveals precision and stability - how much a reading scatters or drifts over time - but it cannot reveal accuracy, because the historian records whatever value the transmitter sends, bias included. A biased instrument produces a clean, steady trend at the wrong value. Only comparing the transmitter against a known reference in a calibration check can establish whether the reading is actually true.
In modern metrology, trueness is closeness to the true value - the absence of bias - while accuracy refers to the overall combination of trueness and precision together. In common usage, though, accuracy is often used loosely to mean closeness to the true value, which is really trueness. The practical point is to separate closeness-to-truth from repeatability, whatever labels you use for them.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.