The F Distribution as a Ratio of Scaled Chi-Square Variables

Why an F Variable Uses Two Scaled Quantities

Some statistical models compare two sources of variation. A natural way to form such a comparison is to put one nonnegative quantity in the numerator and another in the denominator. The F distribution describes a particular kind of ratio: each quantity begins as a chi-square random variable, each is divided by its degrees of freedom, and the two chi-square variables must be independent.

The scaling matters. Dividing each chi-square variable by its own degrees of freedom puts the two quantities on a common scale before taking their ratio. The order also matters: the degrees of freedom attached to the numerator are listed first, and those attached to the denominator second. This tutorial defines that construction and works through how to recognize it; it does not derive the density of the resulting ratio.

Chi-Square Variables and Degrees of Freedom

A chi-square distribution is indexed by its degrees of freedom, written $\nu$. Degrees of freedom may be any positive real number in the distribution’s general definition, not only a positive integer. One way to specify the distribution for any $\nu>0$ is as a gamma distribution with shape $\nu/2$ and scale $2$. Equivalently, its density is

f(x)=\frac{1}{2^{\nu/2}\Gamma(\nu/2)}x^{\nu/2-1}e^{-x/2},\qquad x>0.

Here $\Gamma$ is the gamma function, which supplies the normalizing constant in the density. This definition ensures that a chi-square variable with any positive degrees of freedom takes positive values. Its mean is $\nu$, as follows from the gamma-distribution mean: shape multiplied by scale, $(\nu/2)2=\nu$.

When $\nu$ is a positive integer, there is also a useful representation: if $Z_1,\ldots,Z_\nu$ are independent standard normal random variables, then $Z_1^2+\cdots+Z_\nu^2\sim\chi^2_\nu$. This finite sum representation applies only when $\nu$ is an integer. For noninteger degrees of freedom, the gamma definition applies; a sum with a noninteger number of terms is not the definition.

Definition: Let $U\sim\chi^2_{\nu_1}$ and $V\sim\chi^2_{\nu_2}$ be independent chi-square random variables, with $\nu_1>0$ and $\nu_2>0$. The random variable $$ F=\frac{U/\nu_1}{V/\nu_2} $$ has an F distribution with numerator degrees of freedom $\nu_1$ and denominator degrees of freedom $\nu_2$. We write $F\sim F(\nu_1,\nu_2)$. The parameter order is numerator first, denominator second.

The independence condition is part of this definition, not an optional convenience. The two variables must be independent chi-square variables with the stated degrees of freedom. Also, the numerator is $U/\nu_1$, and the denominator is $V/\nu_2$; dividing by degrees of freedom is part of the construction. Since both chi-square variables are positive, both scaled variables are positive and their ratio is positive.

Worked Examples: Constructing and Reading F Variables

Worked Example: Scaling Two Chi-Square Variables

In a hypothetical calculation, let $U\sim\chi^2_5$ and $V\sim\chi^2_{12}$ be independent. Suppose a particular pair of realized values is $U=8.4$ and $V=10.8$. Define $F=(U/5)/(V/12)$. Find the realized value and identify its distribution.

The stated degrees of freedom match the variables: $U$ has $\nu_1=5$, and $V$ has $\nu_2=12$. Independence is also explicitly assumed, so the definition applies. Scale each value by its own degrees of freedom:

\frac{U}{5}=\frac{8.4}{5}=1.68,\qquad \frac{V}{12}=\frac{10.8}{12}=0.9000.

Now form the ratio in the specified order:

F=\frac{1.68}{0.9000}=1.866\overline{6}\approx 1.867.

Before the values are observed, the random variable has distribution $F\sim F(5,12)$. The realized value is approximately $1.867$. The values $5$ and $12$ identify the distribution; they are not the realized value of $F$.

Worked Example: Noninteger Degrees of Freedom

Suppose $A\sim\chi^2_{2.5}$ and $B\sim\chi^2_{7.5}$ are independent. For a hypothetical realization, let $A=3.0$ and $B=9.0$. Define $R=(A/2.5)/(B/7.5)$. Find $R$ and state its distribution.

Both degrees of freedom are positive, and independence is given. The chi-square distributions are therefore well-defined by the gamma definition, even though their degrees of freedom are not integers. Compute the two scaled values:

\frac{A}{2.5}=\frac{3.0}{2.5}=1.2,\qquad \frac{B}{7.5}=\frac{9.0}{7.5}=1.2.

The ratio is

R=\frac{1.2}{1.2}=1.

Thus $R\sim F(2.5,7.5)$, and this particular realization equals $1$. No finite collection of $2.5$ or $7.5$ standard normal variables is needed or intended here. The positive-real-degree-of-freedom gamma definition is sufficient.

Worked Example: Keeping the Numerator and Denominator Straight

Let $C\sim\chi^2_8$ and $D\sim\chi^2_3$ be independent. A hypothetical calculation defines $Q=(D/3)/(C/8)$. If $C=12$ and $D=6$, find $Q$ and its distribution.

The formula puts $D$, with 3 degrees of freedom, in the numerator; $C$, with 8 degrees of freedom, is in the denominator. The variables are independent, and each is divided by its own degrees of freedom, so the definition applies. Substitute the values:

Q=\frac{6/3}{12/8} =\frac{2}{1.5} =\frac{4}{3} \approx 1.333.

Because the numerator degrees of freedom come first, $Q\sim F(3,8)$. Reversing the parameter order would describe a different arrangement, with the 8-degree-of-freedom quantity in the numerator. The names $C$ and $D$, or the sizes of their observed values, do not determine the order; the formula does.

Why Divide by Degrees of Freedom?

A chi-square variable with $\nu$ degrees of freedom has mean $\nu$. Dividing it by $\nu$ therefore gives a scaled chi-square variable with mean $1$:

E\!\left[\frac{U}{\nu_1}\right]=\frac{E[U]}{\nu_1} =\frac{\nu_1}{\nu_1}=1,\qquad E\!\left[\frac{V}{\nu_2}\right]=\frac{E[V]}{\nu_2} =\frac{\nu_2}{\nu_2}=1.

This explains the role of scaling: it expresses each chi-square quantity relative to its own degrees of freedom, instead of comparing raw totals that may have different degrees of freedom. Both scaled variables have mean $1$, but this does not mean their ratio must equal $1$, nor does it mean the ratio itself has mean $1$. The F variable is the ratio of two random quantities, not a comparison of their means.

A short algebraic rewrite can make the scaling especially clear:

F=\frac{U/\nu_1}{V/\nu_2} =\frac{\nu_2 U}{\nu_1 V}.

This identity follows by multiplying the numerator and denominator of the fraction by $\nu_1\nu_2$. It is useful for checking an expression, but the definition in scaled form is often safer: first divide each chi-square variable by its own degrees of freedom, then divide the numerator result by the denominator result.

Identify the two variables.
Confirm that each quantity is a chi-square random variable, and record its own degrees of freedom.

Check independence.
The numerator and denominator chi-square variables must be independent for the stated F construction.

Scale each variable separately.
Divide each chi-square variable by its own degrees of freedom.

Preserve the order.
Write the numerator degrees of freedom first and the denominator degrees of freedom second in $F(\nu_1,\nu_2)$.

Common Mistakes and Why They Matter

Using raw chi-square values as the F variable. The ratio $U/V$ is generally not the defined F construction. The required ratio is $(U/\nu_1)/(V/\nu_2)$, with each variable divided by its own degrees of freedom.
Reversing the parameter order. In $F(\nu_1,\nu_2)$, $\nu_1$ belongs to the numerator and $\nu_2$ to the denominator. The order follows the formula, not the numerical size of the degrees of freedom.
Ignoring dependence. Two variables having chi-square distributions is not enough: the definition here also requires that they be independent.
Restricting degrees of freedom to integers. A finite sum of squared standard normals represents a chi-square variable only for positive integer degrees of freedom. The gamma definition permits any positive real degrees of freedom.
Assuming the ratio has mean $1$ because both scaled quantities do. Although each scaled chi-square variable has mean $1$, a ratio of random variables does not inherit that mean. Do not infer the mean of the F variable from the separate means without a valid derivation.

Getting the definition right is useful whenever a model leads to a comparison of scaled chi-square quantities. It tells you exactly what distributional family is being used and which degrees of freedom label it. Later calculations involving F probabilities or quantiles depend on identifying those details correctly.

Key takeaway: If $U\sim\chi^2_{\nu_1}$ and $V\sim\chi^2_{\nu_2}$ are independent, then $(U/\nu_1)/(V/\nu_2)\sim F(\nu_1,\nu_2)$. Scale each variable by its own degrees of freedom, and list the numerator degrees of freedom first.

Check Your Understanding

For each question, use the definition of an F variable and keep the numerator and denominator order explicit.

If $X\sim\chi^2_4$ and $Y\sim\chi^2_{9}$ are independent, what distribution does $(X/4)/(Y/9)$ have?
For independent $A\sim\chi^2_6$ and $B\sim\chi^2_2$, identify the distribution of $(B/2)/(A/6)$. Which degrees of freedom come first?
Can a chi-square distribution with $3.5$ degrees of freedom be used in the definition of an F variable? State the definition that supports your answer.
Suppose $U=10$, $\nu_1=5$, $V=6$, and $\nu_2=3$ for independent chi-square variables. Calculate $(U/\nu_1)/(V/\nu_2)$.
Why is independence included in the definition, and why is it not enough to know only the two chi-square distributions?