What happens when λ is zero?

The distribution becomes the central F distribution with the same degrees of freedom.

Not in this standard non-central F parameterization.

Is the mean always defined?

No. Its finite mean requires denominator degrees of freedom greater than two.

AI Glossary / Statistics and probability / Advanced

Non-central F distribution

Definition

The non-central F distribution is the distribution of a ratio formed by dividing an independent noncentral chi-square variable and a central chi-square variable by their respective degrees of freedom. It has numerator and denominator degrees of freedom and a nonnegative noncentrality parameter. It commonly describes F statistics under specified alternatives in statistical models, making it useful for power calculations. When the noncentrality parameter is zero, it reduces to the ordinary central F distribution.

What it means

An F statistic often compares a measure containing a possible systematic effect with a measure of unexplained variation. Under an appropriate null model, both arise from central chi-square quantities. Under a specified alternative, the numerator can contain a systematic component represented by noncentrality.

The distribution therefore helps answer how often a planned test would exceed its rejection threshold if a particular alternative were true. The threshold is ordinarily chosen using the null distribution; the probability of crossing it is then evaluated under the alternative distribution. Those two roles should not be confused.

How it works

Let X have a noncentral chi-square distribution with ν1 degrees of freedom and noncentrality λ, and let Y independently have a central chi-square distribution with ν2 degrees of freedom. Then F = (X/ν1)/(Y/ν2) has the non-central F distribution with those parameters.

For integer ν1, X can be constructed as the sum of squares of independent unit-variance normal variables with means μ_j. In that construction, λ is the sum of μ_j². The denominator uses independent zero-mean unit-variance normal variables. More generally, the distribution is defined for positive real degrees of freedom.

To calculate power, determine the test's degrees of freedom and its model-specific noncentrality under the alternative. Obtain a critical value from the central F distribution and evaluate the noncentral upper-tail probability beyond it. Use a validated numerical implementation, with particular care about extreme tails and parameter conventions.

Three practical examples

A researcher plans an analysis-of-variance experiment and evaluates power under a specified pattern of group means.
An AI evaluation team studies the ability of a regression F test to detect a prespecified set of additional predictors under its modeling assumptions.
A statistics simulation demonstrates how a systematic numerator component changes the distribution of a ratio statistic without simply adding a constant to its values.

Think of it this way

Think of comparing a signal-bearing gauge with a separate noise gauge. A systematic effect changes the statistical behavior of the signal-bearing gauge, while the noise gauge still fluctuates. The ratio's behavior depends on both. Noncentrality summarizes the specified signal in the model; it is not a label saying the observed result is important.

Worked example

Consider an original construction with four numerator normal variables having means 1, 2, 0, and 0, each with variance one. Their squared-mean sum is λ = 1² + 2² = 5. Let the denominator be an independent central chi-square variable with 20 degrees of freedom.

The resulting ratio has ν1 = 4, ν2 = 20, and λ = 5. Its numerator chi-square mean is 4 + 5 = 9, and its denominator chi-square mean is 20. A possible realization X = 12 and Y = 18 gives F = (12/4)/(18/20) = 10/3, approximately 3.333.

The distribution's mean, since ν2 exceeds two, is ν2(ν1 + λ)/(ν1(ν2 − 2)) = 20 × 9/(4 × 18) = 2.5. Dividing the means of the normalized components instead gives (9/4)/(20/20) = 2.25, illustrating that the expectation of a ratio is not generally the ratio of expectations.

With λ = 0 and the same degrees of freedom, the mean is 20/18, approximately 1.111. These calculations describe a distribution, not a p-value or power result; either would also require a specified threshold and testing question.

Illustrative case study

Fictional case: ModelWillow plans a regression study to assess whether several workflow features add explanatory value. Its analysts calculate the appropriate noncentrality for a range of plausible effect scenarios rather than assuming one optimistic effect is certain.

They compare the resulting power curves across sample sizes and check the model's error and design assumptions. When they realize the planned observations are clustered by site, they revisit the calculation rather than treating the original independent-observation F model as automatically applicable. The case illustrates the distribution's planning role and the need to connect its parameters to the actual design.

Common misconceptions and mistakes

Noncentrality is not a location shift. Adding a constant to a central F variable does not generally produce a non-central F variable. Nor is λ the observed F statistic or the probability that the alternative is true.

The numerator and denominator degrees of freedom have different roles and cannot be swapped casually. The ordinary distribution uses a central denominator; making both components noncentral gives a different family. Some moments require sufficiently large denominator degrees of freedom, so a valid distribution need not have every finite moment. Power based on a guessed effect is a scenario calculation, not a guarantee.

Business use

For businesses, the main value is in designing interpretable comparisons before collecting data. A power analysis can show whether a proposed experiment has a reasonable chance of detecting an effect large enough to matter under its assumptions.

Document the model, significance level, effect scenario, sample allocation, degrees of freedom, and noncentrality calculation. Explore several plausible scenarios and distinguish statistical detectability from practical benefit. If the study uses repeated observations, unequal variances, or a complex design, check whether the intended statistic actually follows the assumed noncentral F distribution.

Check your understanding

Is a central F distribution shifted five units to the right the same as a non-central F distribution with λ = 5?

Answer: No. Noncentrality changes the numerator's chi-square distribution and therefore the ratio's shape. It is not an additive location parameter.

Related terms

F distribution
Noncentrality parameter
Degrees of freedom
Analysis of variance
Critical value
noncentral chi-square distribution
statistical power
regression F test

Learn next

Review ratios of independent chi-square variables, then connect noncentrality to the effect and design in a specific ANOVA or regression model. Practice separating null critical values from alternative-distribution power calculations.

Regression model

Sources

SciPy, official noncentral F reference: https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.ncf.html

R, official F distribution documentation: https://stat.ethz.ch/R-manual/R-devel/library/stats/html/Fdist.html

Boost Math, construction and moments: https://www.boost.org/doc/libs/latest/libs/math/doc/html/math_toolkit/dist_ref/dists/nc_f_dist.html

The construction and mean calculations are original and independently checked. ModelWillow is fictional.

Previous: Non-capturing group Next: Non-destructive editing