Measurement uncertainty is frequently represented using probability distributions. A probability distribution describes the relative likelihood of possible values of a quantity. The appropriate distribution depends on the physical process, experimental observations and available information. A normal distribution is commonly used for quantities influenced by many small independent effects. A rectangular distribution may represent limited information in which all values inside a specified interval are treated as equally plausible. A triangular distribution gives greater weight to values near the center of an interval. A uniform distribution is another name for a rectangular distribution in this context. Other distributions, such as binomial, Poisson, exponential, lognormal and Student's t-distributions, may be appropriate for particular measurement processes and statistical analyses. The mean or expectation describes the central value of a distribution, while variance and standard deviation characterize its spread. Statistical models are used to connect observations with unknown parameters and to represent relationships between input quantities and measurement results. A model must be selected carefully because an inappropriate distribution can produce misleading uncertainty estimates. For example, assuming a normal distribution for a strongly asymmetric quantity may be unsuitable. The distribution also affects the interpretation of coverage intervals. A coverage factor cannot be interpreted independently of the distribution and the available degrees of freedom.
Probability Distributions and Statistical Models in Measurement
Probability distributions and statistical models describe how measurement values or uncertainty contributions are expected to vary. They provide the mathematical foundation for evaluating standard uncertainty, combining contributions, estimating intervals and interpreting measurement data.
For a continuous random variable X with probability density function f(x), E(X) = ∫x f(x) dx and Var(X) = ∫(x − μ)² f(x) dx, where μ = E(X).
Explanation
Variables
| Symbol | Meaning | Si Unit |
|---|---|---|
X | Random variable representing a measurement quantity or observation | Unit of the represented quantity |
f(x) | Probability density function | Reciprocal of the unit of x for a continuous quantity |
μ | Expected value or population mean | Same unit as X |
Var(X) | Variance of the random variable | Square of the unit of X |
σ | Population standard deviation | Same unit as X |
Example or application
A worked example has not yet been added.
Definition
A probability distribution is a mathematical description of the probabilities associated with possible values of a random variable. A statistical model is a mathematical representation of the process generating observations or relating data to unknown quantities.
Purpose
To provide mathematical models for describing measurement variation, estimating uncertainty and interpreting the probability associated with possible measurement values.
Importance
Selecting an appropriate probability distribution is essential for reliable Type B evaluation, statistical inference, uncertainty propagation and coverage-interval calculation.
Probability Distribution Fundamentals
Random Variable
A variable whose numerical value is determined by a random process.
Probability Density Function
A function describing the relative distribution of probability for a continuous random variable.
Probability Mass Function
A function assigning probabilities to individual values of a discrete random variable.
Cumulative Distribution Function
The probability that a random variable is less than or equal to a specified value.
Normalization
The total probability across all possible values must equal one.
Common Distributions
| Name | Characteristics | Parameters | Applications | Caution |
|---|---|---|---|---|
Normal distribution | Continuous, symmetric and bell-shaped. It is defined by its mean and standard deviation. |
|
| Not every measurement dataset follows a normal distribution. |
Rectangular or uniform distribution | Every value inside a defined interval has equal probability density. |
|
| Equal plausibility must be supported by the available information. |
Triangular distribution | Probability density increases toward or decreases from a central most-plausible value. |
|
| The shape should reflect actual knowledge rather than convenience. |
Student's t-distribution | Symmetric distribution with heavier tails than the normal distribution, controlled by degrees of freedom. |
|
| Its use depends on the statistical model and assumptions. |
Binomial distribution | Discrete distribution describing the number of successes in a fixed number of independent trials with constant success probability. |
|
| Trials must satisfy the model's assumptions. |
Poisson distribution | Discrete distribution used for counts of events occurring independently at a constant average rate under appropriate conditions. |
|
| The assumed event process must be appropriate. |
Lognormal distribution | A positive-valued, right-skewed distribution whose logarithm is normally distributed. |
|
| A symmetric normal model may be inappropriate for strongly skewed positive data. |
Statistical Model Components
Observation Model
Describes how measured observations relate to the underlying quantity and random effects.
Parameter Model
Describes unknown parameters such as a population mean, variance or calibration coefficient.
Measurement Model
Describes the mathematical relationship between input quantities and the final measurement result.
Residual Model
Describes the differences between observed data and model predictions.
Distribution Selection
Identify whether the quantity is continuous or discrete.
Examine the physical mechanism generating the variation.
Inspect repeated observations when available.
Consider symmetry, skewness, bounds and possible outliers.
Review the source of Type B information.
Select a distribution supported by the evidence.
Document the assumptions and evaluate whether the model is adequate.
Worked Example
Problem
A measurement contribution is known to lie between −0.03 mm and +0.03 mm, and all values within this interval are considered equally plausible.
Distribution
Rectangular distribution
Calculation
u = 0.03/√3 ≈ 0.0173 mm
Result
The standard uncertainty is approximately 0.017 mm.
Interpretation
The probability model converts the known interval into a standard deviation-like uncertainty value.
Model Validation
Inspect residual patterns for systematic structure.
Check whether the assumed distribution is consistent with available data.
Investigate dependence between observations.
Examine whether the model covers all significant physical influences.
Assess whether the model is appropriate over the full measurement range.
Use sensitivity analysis to identify influential assumptions.
Limitations
A probability distribution is a model, not proof of the exact behavior of a measurement process.
Limited data may not reliably reveal distribution shape.
Incorrect assumptions can produce misleading coverage intervals.
Correlations between quantities must be considered where relevant.
A normal approximation may be unsuitable for small samples or strongly asymmetric results.
Reference
JCGM 100:2008, Evaluation of Measurement Data — Guide to the Expression of Uncertainty in Measurement (GUM).
NIST Technical Note 1297, Guidelines for Evaluating and Expressing the Uncertainty of NIST Measurement Results.
JCGM 101:2008, Evaluation of Measurement Data — Supplement 1 to the GUM: Propagation of Distributions Using a Monte Carlo Method.
Related Topic
SCI-REF-003-083
SCI-REF-003-085
SCI-REF-003-086
SCI-REF-003-087
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