Relative Error Calculator

An error of 0.09 means nothing until you know what it is an error in — negligible against the distance to the Moon, routine in a school laboratory, and utterly disqualifying in a machined shaft or a pharmaceutical assay. This calculator supplies the missing context in full: enter the measured value and the true (accepted) value, and it returns the absolute error, the relative error and the percent error — the trio every lab report, calibration certificate, sensor datasheet and quality record is ultimately built from.

Three Errors, One Ladder

The three outputs are a ladder of increasing context, each rung answering the question the rung below it cannot answer alone. Absolute error = |measured − true|: the raw distance, in the quantity’s own units, indispensable and uninterpretable alone. Relative error divides that distance by the true value: a pure ratio, unitless, comparable across quantities — 0.0092 says the miss is about a hundred-and-ninth of the thing measured, whether the thing was grams or light-years. Percent error is the relative error wearing the percent sign (× 100), the form in which tolerances, datasheets and grading rubrics quote it. The ladder’s direction matters professionally: work down from percent to absolute when setting specifications (a 0.5% tolerance on a 200 mm part is a 1 mm allowance), and up from absolute to percent when reporting results, where comparability is the point.

Worked Example: Measuring Gravity in the Teaching Lab

The defaults are the classic first-year experiment: a pendulum rig measures g as 9.72 m/s² against the accepted 9.81. Absolute error: |9.72 − 9.81| = 0.09 m/s². Relative error: 0.09 ÷ 9.81 = 0.00917. Percent error: 0.92% — a respectable teaching-lab result (sub-1% with string and a stopwatch is good work; the signed version, −0.92%, adds that the measurement ran low, sending the thoughtful student hunting for a systematic cause — a stretched string, a timing bias — rather than random noise). Watch the ladder do its job: 0.09 alone could be a catastrophe or a triumph; 0.92% settles the question in one glance, for this experiment and for a chemist’s titration or a surveyor’s traverse judged by the same figure.

Percent Error Versus Percent Difference

The family’s most-confused cousins deserve their paragraph. Percent error (this page) requires a true value — a standard, a reference, an accepted constant — and measures one value’s distance from it. Percent difference compares two measurements, neither privileged, dividing by their average: |a − b| ÷ ((a+b)/2) × 100. Two students measuring g as 9.72 and 9.86 differ by 0.14 ÷ 9.79 ≈ 1.43% (difference), while their errors against 9.81 are 0.92% and 0.51% respectively. Choosing wrong is not pedantry: inter-lab comparisons, replicate agreement and method validation live on percent difference, while accuracy claims against standards live on percent error — and a report that swaps them has quietly changed what it is claiming.

What Counts as True

The formula’s innocence conceals a philosophical input: the “true” value is whatever the exercise agrees to worship — the CODATA constant, the certified reference material, the calibrated standard weight, the theoretical prediction. Consequences follow. Percent error can never be smaller than the true value’s own uncertainty permits you to know (a school balance checked against a weight certified only to 1% cannot certify 0.1% work); against a predicted value, “error” quietly becomes “discrepancy”, carrying model failure among its suspects; and a true value of exactly zero makes the ratio explode — the calculator refuses it explicitly, because relative error against zero is not a large number but a category mistake, where absolute error must carry the report alone.

Reporting the Error: Significant Figures

Errors obey the same decimal discipline as uncertainties: the percent error is quoted to two significant figures at most (“0.92%”, not “0.917431%”), because the measured value itself rarely justifies more, and spurious precision in an error figure is a small fraud against the reader. The absolute error inherits the measured value’s decimal places (0.09, matching 9.72’s hundredths), and a well-kept lab book shows the trio — measured, true, percent error — at consistent precision, so a checker can recompute any one from the other two. This calculator prints generously and expects you to round for publication: the arithmetic’s job is to be right, the report’s job is to be honest about how right.

Calibration Chains: Where True Values Grow Up

Follow a “true value” backwards and it dissolves into earlier measurements: the lab’s reference weight was certified against a national standard, itself compared to the international prototype’s descendants — a chain of comparisons in which each link carries its own relative error, and the final “truth” your measurement is judged against is really the chain’s last, best number. Metrologists call the property traceability, and its arithmetic is this page’s: each link’s uncertainty combines (roughly in quadrature) with the next, so a school measurement’s 0.92% against a certified value certified to 0.01% is meaningfully 0.92%, while the same figure against a value known only to 0.5% must be reported with the humility the chain imposes. Error is always measured against a reference; wisdom is knowing the reference’s own biography.

Random Scatter, Systematic Drift: The Error’s Two Tempers

A single percent error cannot say why the measurement missed, but a series of signed errors can, and the distinction governs what to fix. Suppose four pendulum trials return signed errors of +0.5%, −0.9% (the worked example), +0.3%, −0.7%. The scatter’s size — root-mean-square ≈ 0.64% — prices the method’s precision; the mean signed error, −0.2%, whispers a systematic lean (three of four negative, all the large ones negative) pointing at the apparatus rather than luck: a mismeasured string length, a consistent stopwatch anticipation. Random error is beaten by repetition and averaging; systematic error is immune to both and yields only to calibration and redesign — which is why the calculator’s note preserves the sign the headline definition discards. The direction of misses is the diagnostic; the magnitude is merely the measurement.

Targets by Trade, Once More With Numbers

To fix the “good error” question in the pocket: an analytical balance’s check weights are verified to about 0.01%; a surveyor’s closed traverse is typically accepted under roughly 0.02–0.1% of its length (relative precisions of 1:5,000 to 1:50,000 in the trade’s own dialect); a CNC machine holds parts to thousandths that work out near 0.005% of a 200 mm dimension; and the teaching laboratory celebrates anything under 1%. The same formula spans five orders of magnitude of ambition — which is the deepest reason the relative form, not the absolute, is the one worth standardising a page around.

Related Tools

More from the Statistics Calculators hub: the Absolute Uncertainty Calculator — error’s forward-looking twin — and the Error Propagation Calculator for carrying errors through calculations.

Frequently Asked Questions

What is the formula for relative error?

Relative error = |measured − true| ÷ |true|, with percent error the same ratio times 100. The worked example runs it end to end: 0.09 ÷ 9.81 = 0.00917, or 0.92%. The absolute value in the standard definition strips direction; keep the sign (the calculator’s note does) when the direction is diagnostic, as it is in calibration and bias hunting.

Is percent error the same as relative error?

Numerically one is the other times a hundred; rhetorically they serve slightly different rooms — relative error (the decimal ratio) is the scientist’s internal currency, percent error the public and specification form. They are never in disagreement, and converting between them costs nothing, which is why this calculator prints both rungs of the ladder plus the absolute error beneath them.

Can percent error be over 100%?

Yes, and the cases are instructive: measuring 25 against a true 10 is a 150% error. It startles students raised on percentages that cap at 100, but error is a ratio to the truth, not a share of a whole — overshooting by more than the quantity itself is perfectly expressible. (Under-reading, symmetrically, can never exceed 100% without passing through zero into negative quantities, an asymmetry built into the ratio’s definition.)

What is a good percent error?

Entirely jurisdiction-dependent: under 1% is excellent teaching-lab work and routine analytical chemistry; 5% passes many school rubrics; engineering acceptance runs to the tolerance on the drawing (often far tighter); clinical and metrology contexts live at fractions of a percent with traceable standards. The number’s virtue is that it lets each field state its bar in one figure — this calculator’s job is only to ensure the figure is computed the standard way, every time, without exception or improvisation.

Why does my percent error use the experimental value in the denominator?

It shouldn’t — that variant circulates in school notes but the standard definition divides by the true (accepted) value, because the truth is the reference the error is an error from. Dividing by the measured value answers a different, rarely asked question and inflates or deflates the figure whenever the measurement strays. This calculator follows the standard; if a rubric demands the other, divide the absolute error by your measured value by hand and note the deviation in the report.

How do errors in two measurements combine?

That is propagation’s territory, one hub away: sums and differences add absolute errors, products and quotients add relative errors (approximately, for independent random errors, in quadrature). The relative errors this page computes are exactly the currency those combinations spend — one more reason the ratio, rather than the raw difference, is the form worth reporting.