Relative Error Calculator

Absolute error is the raw difference between the measured value and the true value: |measured − true|. It is expressed in the same units as the measurement. Relative error is this difference divided by the true value: |measured − true| / |true|, giving a unitless fraction. Percentage error multiplies relative error by 100. For example, if a scale reads 10.3 g when the true mass is 10.0 g, the absolute error is 0.3 g, the relative error is 0.03, and the percentage error is 3%.

Acceptable percentage error depends entirely on the context. In analytical chemistry, errors below 1% are expected. In physics lab experiments, 5% is often acceptable. Engineering measurements may require errors below 0.1%. Survey and social science data may accept 10% or more. There is no universal threshold - the key question is whether the error is small enough for the decision or conclusion that depends on the measurement.

Both RMSE and MAE measure average prediction error across a set of forecasts or model outputs. MAE (Mean Absolute Error) gives equal weight to all errors and is easier to interpret: an MAE of 5 means predictions are off by 5 on average. RMSE (Root Mean Square Error) squares the errors before averaging, which gives extra weight to large errors. RMSE ≥ MAE always. Use MAE when all errors are equally important; use RMSE when large errors are disproportionately costly (e.g. in safety-critical forecasting).

The relative error formula divides by the true value because we are measuring how far the measured value deviates from what is correct. The true value is the reference standard. If you divided by the measured value instead (which is the relative difference from the other direction), you would get a different number - and crucially, if the measured value is much smaller than the true value, dividing by it would artificially inflate the error. Always divide by the true or accepted reference value.

Yes. If the measured value is more than double the true value, or if it has the opposite sign, the percentage error exceeds 100%. For example, if the true value is 2 and the measured value is 10, the percentage error is |10−2|/2 × 100% = 400%. This most commonly occurs with very small true values or when there is a systematic instrument error or incorrect calculation.

Relative deviation from the mean is |x − x̄| / x̄, where x̄ is the mean of a set of measurements. Unlike the standard relative error, it does not require a known true value - instead, the mean serves as the reference. It is used when you want to assess the consistency of repeated measurements or values within a group, for example comparing how far each measurement deviates from the average in a laboratory experiment where the true value is unknown.

RMSE is calculated as: √(Σ(predicted_i − actual_i)² / n). First, compute the squared error for each prediction: (predicted − actual)². Sum all squared errors, divide by the number of observations n, then take the square root. RMSE has the same units as the original values, making it more interpretable than MSE (mean squared error). A lower RMSE indicates better model accuracy.

When RMSE >> MAE, it signals that there are a few predictions with very large errors pulling the RMSE up, while most predictions are reasonably accurate. This gap between RMSE and MAE is diagnostic: a large difference indicates the presence of outlier predictions or systematic errors in specific cases. For a model with uniformly distributed errors, RMSE is only modestly larger than MAE (roughly 1.25× for normally distributed errors).