standarddeviationcalculator.net

Updated

Error and uncertainty calculators

Calculators for how far a measurement, estimate or prediction is from the truth, and for carrying that uncertainty through a calculation. Lab students, engineers and anyone checking a forecast will find the main tools here, along with the entropy and diversity indices used in ecology and information theory.

Which calculator do I need?

You have or wantUse
Several measured values with ± uncertainties, combined in a formulaError propagation calculator
Repeated readings of the same quantity and you need its uncertaintyUncertainty calculator
A ± value to switch between absolute and percentage formAbsolute uncertainty calculator
One measured value against a known true or accepted valueRelative error calculator
Actual and predicted values from a model or forecastMSE calculator
How precise an estimate is relative to its own sizeRelative standard error calculator
Species counts from a survey and you want a diversity indexSimpson's diversity index

Measurement uncertainty

Quantify the uncertainty in lab readings and carry it through to a final result.

Prediction and estimate error

How far a model misses the data, and how precise an estimate is.

Standard scores

Move between a raw value and its position in a distribution.

Diversity and information

Indices that measure how evenly counts are spread across categories.

Absolute, relative and propagated error

An absolute uncertainty is in the units of the measurement: 10.0 ± 0.2 cm. A relative uncertainty divides that by the value (0.02, or 2%), which lets you compare the precision of quantities measured in different units. Relative error is the same idea applied to a known true value: measuring 9.8 when the answer is 10 is an absolute error of 0.2 and a relative error of 2%.

When measurements are combined, their uncertainties combine too. Take A = 10.0 ± 0.2 and B = 5.0 ± 0.1, independent. For the sum, absolute uncertainties add in quadrature: A + B = 15.0 ± √(0.2² + 0.1²) = 15.0 ± 0.22. For the product, relative uncertainties add in quadrature: both are 2%, so A × B = 50 ± 50 × √(0.02² + 0.02²) = 50 ± 1.4. Simply adding the uncertainties would give ±0.3 and ±2.0, the worst case, which overstates the typical error. The error propagation calculator applies the general rule to any formula.

Measuring prediction error

A residual is one observed value minus its prediction. The MSE calculator summarises a whole set: for errors of 1, −2 and 3, the mean absolute error is 2, the mean squared error is (1 + 4 + 9) / 3 = 4.67 and the root mean squared error is 2.16. RMSE is in the original units and weights large misses more heavily than MAE does, so the gap between the two hints at occasional big errors.

Common mix-ups

  • Standard uncertainty is not the SD of the readings. For the mean of n repeated readings, the standard uncertainty is s / √n, as the uncertainty calculator shows.
  • Relative standard error is not relative standard deviation. RSE divides a standard error by its estimate and tells you how reliable that estimate is; %RSD divides the SD by the mean and describes the spread of the data.
  • Simpson's D and 1 − D point in opposite directions. D is the chance two individuals picked at random are the same species, so a higher D means less diversity. The calculator reports D, 1 − D and 1/D side by side.

Guides to read alongside

Common questions

How many significant figures should an uncertainty have?

Usually one, or two if the first digit is 1. Then round the measured value to the same decimal place as the uncertainty, so 9.8134 ± 0.0237 becomes 9.81 ± 0.02.

Should I use MSE, RMSE or MAE to compare models?

RMSE is the usual choice because it is in the original units and punishes large errors. Use MAE when occasional large errors should not dominate the comparison, and MAPE when you need a percentage that works across series of different sizes, as long as no actual value is zero.

What is the difference between Shannon and Simpson diversity?

Both rise with the number of species and how evenly individuals are spread among them. Shannon's H′ gives more weight to rare species; Simpson's index is driven mainly by the most common ones. Report whichever your field or course uses, and do not compare values of one with the other.