Many measurements are calculated indirectly from multiple input quantities. For example, density is calculated from mass divided by volume, and electrical resistance may be determined from voltage and current. Each input quantity has an associated uncertainty, and these uncertainties influence the uncertainty of the final result. The relationship is described by a measurement model, y = f(x₁, x₂, ..., xₙ). The sensitivity coefficient for an input quantity describes how strongly the output changes when that input changes. For sufficiently small uncertainties and a sufficiently smooth measurement model, the law of propagation of uncertainty provides an approximate method for calculating combined standard uncertainty. If the input quantities are uncorrelated, the squared contributions are added. If they are correlated, covariance terms must also be included. This is important because shared calibration standards, common environmental influences and common data-processing steps can produce correlations. Combined standard uncertainty is not obtained by simply adding all input uncertainties. Each contribution must first be expressed in a compatible form and transformed through the measurement model. When the model is strongly nonlinear, input distributions are non-normal or the uncertainty is large relative to the model's behavior, numerical methods such as Monte Carlo propagation may be more appropriate.
Combined Standard Uncertainty and Uncertainty Propagation
Combined standard uncertainty is the standard uncertainty of a measurement result calculated from the uncertainty contributions of its input quantities. Uncertainty propagation uses the measurement model and sensitivity coefficients to determine how uncertainty in individual inputs influences the final result.
For y = f(x₁, ..., xₙ), u_c²(y) = Σ[cᵢ²u²(xᵢ)] + 2Σᵢ<ⱼ[cᵢcⱼ cov(xᵢ,xⱼ)], where cᵢ = ∂f/∂xᵢ. For uncorrelated inputs, the covariance terms are zero.
Explanation
Variables
| Symbol | Meaning | Si Unit |
|---|---|---|
y | Output quantity or measurement result | Unit of the output quantity |
xᵢ | Input quantity i | Unit of input quantity i |
u(xᵢ) | Standard uncertainty of input quantity i | Same unit as xᵢ |
cᵢ | Sensitivity coefficient for input quantity i | Unit of y divided by unit of xᵢ |
u_c(y) | Combined standard uncertainty of the output | Same unit as y |
cov(xᵢ,xⱼ) | Covariance between input quantities i and j | Product of their respective units |
Example or application
A worked example has not yet been added.
Definition
Combined standard uncertainty is the standard uncertainty of a measurement result obtained by combining the standard uncertainties of the input quantities using the measurement model and their correlations.
Purpose
To determine how uncertainties in individual input quantities contribute to the uncertainty of a calculated measurement result.
Importance
Uncertainty propagation is essential for indirect measurements, engineering calculations, physical constants, calibration equations and scientific models involving multiple measured quantities.
Sensitivity Coefficients
Description
A sensitivity coefficient measures the local rate of change of the output quantity with respect to an input quantity.
Calculation
cᵢ = ∂f/∂xᵢ
Interpretation
A large sensitivity coefficient means that a small change in the corresponding input may produce a relatively large change in the output.
Uncertainty Propagation Rules
| Relationship | Uncertainty |
|---|---|
Addition: y = a + b | For uncorrelated inputs, u_c(y) = √[u²(a) + u²(b)]. |
Subtraction: y = a − b | For uncorrelated inputs, u_c(y) = √[u²(a) + u²(b)]. |
Multiplication: y = ab | For uncorrelated inputs, the relative variance is approximately u_c²(y)/y² = u²(a)/a² + u²(b)/b². |
Division: y = a/b | For uncorrelated inputs, the relative variance is approximately u_c²(y)/y² = u²(a)/a² + u²(b)/b². |
Power: y = aⁿ | For small uncertainty and nonzero a, u_c(y)/|y| ≈ |n|u(a)/|a|. |
Correlation And Covariance
Description
Correlation describes statistical association between input quantities. Covariance represents the direction and magnitude of their joint variation.
Positive Correlation
When inputs tend to increase together, the covariance is positive.
Negative Correlation
When one input tends to increase as the other decreases, the covariance is negative.
Zero Covariance
Zero covariance means no linear covariance, but it does not always establish statistical independence.
Importance
Ignoring significant correlations may either overestimate or underestimate combined uncertainty.
Worked Example
Problem
A rectangle has measured length 2.00 m with standard uncertainty 0.01 m and width 1.00 m with standard uncertainty 0.02 m. Assume the measurements are uncorrelated. Calculate the area and combined standard uncertainty.
Measurement Model
A = L × W
Area
A = 2.00 × 1.00 = 2.00 m²
Sensitivity Coefficients
Length
∂A/∂L = W = 1.00 m
Width
∂A/∂W = L = 2.00 m
Calculation
u_c(A) = √[(1.00 × 0.01)² + (2.00 × 0.02)²] = √0.0017 ≈ 0.0412 m²
Result
A = 2.00 m² with combined standard uncertainty approximately 0.041 m².
Interpretation
Both input uncertainties contribute to the area uncertainty. The width uncertainty contributes more strongly because its sensitivity coefficient is larger.
Uncertainty Propagation Methods
Analytical Method
Uses derivatives, sensitivity coefficients and covariance information. It is commonly used for smooth measurement models and sufficiently small uncertainties.
Numerical Method
Uses numerical evaluation of the measurement model to estimate sensitivity coefficients or propagate uncertainty.
Monte Carlo Method
Samples input quantities from their assigned probability distributions, evaluates the measurement model repeatedly and estimates the output distribution from the simulated results.
Assumptions And Limitations
The first-order propagation law is an approximation and depends on the measurement model's behavior.
Sensitivity coefficients should be evaluated at appropriate estimates of the input quantities.
Correlations must be included where relevant.
Input uncertainties must be expressed in compatible units and statistical forms.
Strong nonlinearities may require higher-order or numerical methods.
A result close to zero may make relative uncertainty difficult to interpret.
Practical Procedure
Write the measurement model.
Identify every input quantity.
Determine the standard uncertainty of each input.
Calculate the sensitivity coefficient for each input.
Determine correlations or covariances.
Calculate the combined uncertainty.
Check units and numerical consistency.
Evaluate whether first-order propagation is appropriate.
Reference
JCGM 100:2008, Evaluation of Measurement Data — Guide to the Expression of Uncertainty in Measurement (GUM).
JCGM 101:2008, Evaluation of Measurement Data — Supplement 1 to the GUM: Propagation of Distributions Using a Monte Carlo Method.
NIST Technical Note 1297, Guidelines for Evaluating and Expressing the Uncertainty of NIST Measurement Results.
Related Topic
SCI-REF-003-081
SCI-REF-003-083
SCI-REF-003-084
SCI-REF-003-087
SCI-REF-003-088
Sources and verification
Sources are shown as provided in the project data. Not every source has yet been independently verified. Please report errors or missing citations.
- Source verification pending; check an authoritative textbook, university, standards body, or peer-reviewed source.
Suggest a correction or source · More in Units & Measurement