Precision vs Accuracy in Measurement: Repeatability Is Not the Same as Truth

Precision is how tightly your measurements agree with each other, and accuracy is how close they land to the true value, so a device can be highly precise and still be reliably wrong. Precision is repeatability. Accuracy is closeness to truth. The two describe different failures, and a number can score well on one while failing badly on the other. Confusing them is a common way a measurement misleads a careful person. This is general education, not medical advice, and any decision about your own care belongs with a qualified clinician.

The picture I keep in mind is a target. Accuracy is whether your shots land near the bullseye. Precision is whether your shots land near each other, wherever they land. Tight and centered is what everyone wants. But tight and off to the side is the dangerous case, because the consistency looks like trustworthiness while the whole cluster sits away from the truth.

What precision and accuracy each measure

Precision answers a question about the instrument talking to itself. Measure the same thing several times under the same conditions, and precision is how little the answers scatter. A precise scale gives you nearly the same reading five times in a row. It says nothing yet about whether that reading is correct.

Accuracy answers a question about the instrument talking to reality. It asks whether the measurements, on average, sit at the true value. An accurate scale reads your true weight, even if it wobbles a little around it from reading to reading.

The reason the words get swapped is that in everyday speech both just mean "good." In measurement they mean two separate things, and the separation is the entire point. A method can be repeatable without being correct, and it can be correct on average while being too noisy to trust for a single reading.

The four combinations, and which one fools you

Put the two properties on the target and four cases appear. Only one is what you want, and only one is a quiet trap.

Accurate and precise is the goal. The shots cluster, and the cluster sits on the bullseye. You can trust a single reading and you can trust that repeats will agree.

Accurate but imprecise is honest but noisy. The shots scatter widely, yet they scatter around the truth, so averaging many of them lands you close. One reading alone might mislead, but the method is not lying to you in a fixed direction.

The trap is precise but inaccurate. The shots land in a tight knot, but the knot sits away from the bullseye. Every reading agrees with the last, which feels like reliability, so the error hides behind the consistency. This is the case that costs people the most, because the instrument looks confident and calm while it is steadily wrong.

Imprecise and inaccurate is at least obvious. The shots are scattered and off target, so nobody is tempted to trust it.

Two different kinds of error underneath

Precision and accuracy separate because measurement error comes in two flavors, and each attacks a different property.

Random error is the jitter. It pushes readings up and down unpredictably from one measurement to the next, and it is what precision measures. Random error averages out if you repeat enough times, which is why pooling several readings tightens a noisy but honest measurement.

Systematic error is bias. It pushes every reading in the same direction by a similar amount, a scale that always reads two kilograms high, an assay calibrated against the wrong standard. This is what accuracy measures, and it is the more insidious of the two, because repeating the measurement does not help. Average a thousand biased readings and you get a very precise estimate of the wrong number.

That last sentence is the whole warning. More data fixes random error. More data only sharpens systematic error, engraving the wrong answer more confidently.

Why the confusion misleads in medicine and research

In clinical work the trap shows up whenever a device is consistent enough to feel trustworthy. A home monitor that reads the same value every morning feels dependable, but if it carries a fixed offset it is precisely misinforming someone about their own body, day after day. The steadiness is what makes the bias hard to notice.

In research the same trap wears a lab coat. A tightly clustered set of measurements produces a small standard deviation and a narrow confidence interval, and both look like signs of a strong result. Neither one detects bias. A confidence interval describes precision, the spread of your estimate, and it can be gorgeously narrow around a value that a calibration problem pushed off the truth. A study can report an effect with great apparent certainty while a systematic error quietly relocated the entire finding.

Some of my own work looked at how insulin sensitivity and insulin response differ across ethnic groups, drawing on a meta-analysis published in Diabetes Care. Pooling studies buys you precision, a tighter estimate from more data. It does nothing about a bias shared across the studies, such as a measurement method that reads consistently differently in one group than another. Precision and accuracy have to be argued separately, and confusing them lets a shared bias survive the pooling untouched.

How to tell them apart in practice

The habit worth building is to ask two questions of any number, never just one.

First, how much would this reading move if I measured again right now? That is a precision question, and you answer it by repeating and watching the scatter. Second, what am I comparing this instrument against to know it is right? That is an accuracy question, and you can only answer it with an external reference: a known standard, a trusted method, ground truth of some kind. Precision you can check by yourself. Accuracy you can only check against something outside the instrument.

When I evaluate a measurement claim, the tell is a report full of tight repeatability with nothing said about the reference standard. That combination should raise suspicion rather than confidence, because it is precisely the profile of the precise-but-wrong case. A single trustworthy number needs both properties, and it earns them in two different ways.

The mental model to carry is short. Precision is agreement with yourself. Accuracy is agreement with the truth. You need both, they break for different reasons, and the most dangerous measurement is the one so consistent that no one thought to check whether it was right.