Why Degrees of Freedom Matter
An F variable is a ratio of two independently scaled chi-square variables. Its degrees of freedom do more than label the distribution: they determine how the density behaves near zero, where its peak lies, and how slowly its right tail decays. Those features matter when interpreting a graph or selecting the correct F distribution for a calculation.
The discussion here is about the shape of the density, not the calculation of F probabilities or moments. Throughout, let the numerator and denominator degrees of freedom be positive numbers \(\nu_1\) and \(\nu_2\), respectively. In the usual sample-variance setting they are positive integers.
The ratio is positive because both chi-square variables are positive with probability 1. The support of an F density is therefore \(x>0\); negative values are impossible. To see how the density’s shape follows from the degrees of freedom, it is useful to write its formula in terms of \(a=\nu_1/2\) and \(b=\nu_2/2\).
Behavior Near Zero and in the Right Tail
Near zero, the factor involving \(1+(\nu_1/\nu_2)x\) approaches 1. The density’s leading behavior is therefore proportional to \(x^{\nu_1/2-1}\). This gives a direct way to classify its left endpoint:
- If \(\nu_1<2\), the density increases without bound as \(x\) approaches zero from above.
- If \(\nu_1=2\), the density approaches a finite, positive value at zero.
- If \(\nu_1>2\), the density approaches zero at zero.
These statements describe the limiting density at the endpoint; the random variable itself still takes values strictly greater than zero. In particular, a density that becomes unbounded at zero does not mean there is positive probability that \(F=0\).
For large \(x\), the quantity \(1+(\nu_1/\nu_2)x\) grows in proportion to \(x\). Combining its power with the factor \(x^{a-1}\) gives a density proportional to \(x^{-b-1}\), or \(x^{-(\nu_2/2+1)}\). Thus the denominator degrees of freedom determine the power governing the far-right tail. A smaller \(\nu_2\) gives a more slowly decaying density tail; a larger \(\nu_2\) gives a faster-decaying one.
Finding the Mode
The mode is the point where a density reaches its maximum. When \(\nu_1>2\), the F density rises from zero, reaches a single interior maximum, and then decreases. Its mode is
When \(\nu_1=2\), the density is greatest at the left boundary in the limiting sense and decreases for positive \(x\). When \(\nu_1<2\), it decreases from an unbounded density at the left boundary. In these cases there is no positive interior mode.
For \(x>0\), taking the logarithm separates the variable terms into \((a-1)\ln x-(a+b)\ln(1+cx)\), where \(c=\nu_1/\nu_2\).
The derivative is \((a-1)/x-(a+b)c/(1+cx)\). Equating it to zero gives \((a-1)(1+cx)=(a+b)cx\).
Rearranging gives \(x=(a-1)/(c(b+1))\), which is positive exactly when \(a>1\), or \(\nu_1>2\). Substituting \(a=\nu_1/2\), \(b=\nu_2/2\), and \(c=\nu_1/\nu_2\) yields the stated mode formula.
This derivation also explains why the condition \(\nu_1>2\) matters. If \(a\leq 1\), the derivative of the log density is negative for every positive \(x\), so there is no positive interior maximum. For \(\nu_1>2\), the density is zero at the left endpoint and tends to zero in the right tail; the single stationary point is its maximum.
Worked Examples: Reading the Shape from the Degrees of Freedom
Worked Example: Classifying the Endpoint and Finding the Mode
Let \(F\sim F_{2,4}\). Identify its support, describe its density near zero and in the far-right tail, and determine whether it has a positive interior mode. Use the F density to verify the endpoint behavior.
Method. Read the endpoint behavior from \(\nu_1\), the tail power from \(\nu_2\), and check the condition for an interior mode.
Here \(\nu_1=2\) and \(\nu_2=4\), so \(a=1\), \(b=2\), and \(\nu_1/\nu_2=1/2\). The support is \(x>0\).
For these parameters, \(B(1,2)=\Gamma(1)\Gamma(2)/\Gamma(3)=1\cdot1/2=1/2\). The density is \(f(x)=\dfrac{(1/2)^1}{1/2}x^0(1+x/2)^{-3}=(1+x/2)^{-3}\), for \(x>0\).
As \(x\) approaches zero, \(f(x)\) approaches \((1+0)^{-3}=1\), a finite positive value. Since \(\nu_1=2\), the mode formula for an interior mode does not apply; the density decreases from its boundary value for every \(x>0\). There is no positive interior mode.
For large \(x\), the density has a tail proportional to \(x^{-(\nu_2/2+1)}=x^{-3}\). This agrees with the density formula, because \((1+x/2)^{-3}\) is proportional to \(x^{-3}\) for large \(x\).
Check. The density is positive only for \(x>0\), approaches 1 at the left endpoint, decreases as \(x\) increases, and has the predicted \(x^{-3}\) right-tail power. The common error to avoid is calling the boundary maximum an interior mode.
Worked Example: Calculating an Interior Mode and Density Height
Let \(F\sim F_{6,4}\). Find its mode and calculate the density at the mode. State the behavior at zero and name the common error this calculation avoids.
Method. Since \(\nu_1=6>2\), use the interior-mode formula. Then substitute the resulting \(x\) into the density.
The degrees of freedom are \(\nu_1=6\) and \(\nu_2=4\). The mode is \(\dfrac{6-2}{6}\dfrac{4}{4+2}=\dfrac{4}{6}\dfrac{4}{6}=\dfrac{16}{36}=\dfrac{4}{9}\).
For the density, \(a=3\), \(b=2\), and \(\nu_1/\nu_2=3/2\). Also, \(B(3,2)=\Gamma(3)\Gamma(2)/\Gamma(5)=2\cdot1/24=1/12\). Thus the density coefficient is \((3/2)^3/(1/12)=(27/8)\cdot12=81/2\), and \(f(x)=\dfrac{81}{2}x^2(1+3x/2)^{-5}\).
At \(x=4/9\), the factor \(1+3x/2\) is \(1+(3/2)(4/9)=1+2/3=5/3\). Therefore, \(f(4/9)=\dfrac{81}{2}\cdot\dfrac{16}{81}\cdot(5/3)^{-5} =8\cdot(3/5)^5=8\cdot243/3125=1944/3125=0.62208\).
Since \(\nu_1>2\), the density approaches zero at zero and has an interior peak at \(4/9\). The density height is \(0.62208\), or \(0.6221\) to four decimal places.
Check. The mode is positive, as required when \(\nu_1>2\); it is below 1, and substitution into the density gives a positive height. The common error to avoid is swapping the degrees of freedom in the mode formula.
Worked Example: Comparing Right Tails with the Same Numerator Degrees of Freedom
Compare \(F_A\sim F_{4,2}\) and \(F_B\sim F_{4,8}\). For each distribution, find the mode and density at \(x=1\). Then compare their far-right tail powers and identify what the comparison shows.
Method. Keep the numerator degrees of freedom fixed, calculate each mode and density value from its own parameters, and use the denominator degrees of freedom to compare the tail powers.
Both distributions have \(\nu_1=4\), so each has an interior mode. For \(F_A\), \(\nu_2=2\), giving \(\operatorname{mode}(F_A)=\dfrac{4-2}{4}\dfrac{2}{2+2}=\dfrac12\cdot\dfrac12=\dfrac14\). For \(F_B\), \(\nu_2=8\), giving \(\operatorname{mode}(F_B)=\dfrac{2}{4}\dfrac{8}{10}=\dfrac12\cdot\dfrac45=\dfrac25\).
For \(F_A\), \(a=2\), \(b=1\), and \(\nu_1/\nu_2=2\). Since \(B(2,1)=\Gamma(2)\Gamma(1)/\Gamma(3)=1/2\), its density is \(f_A(x)=2^2/(1/2)\,x(1+2x)^{-3}=8x(1+2x)^{-3}\). Thus \(f_A(1)=8(1)(3)^{-3}=8/27=0.2963\) to four decimal places.
For \(F_B\), \(a=2\), \(b=4\), and \(\nu_1/\nu_2=1/2\). Here \(B(2,4)=\Gamma(2)\Gamma(4)/\Gamma(6)=1\cdot6/120=1/20\). Its density is \(f_B(x)=(1/2)^2/(1/20)\,x(1+x/2)^{-6}=5x(1+x/2)^{-6}\). Therefore, \(f_B(1)=5(1)(3/2)^{-6}=5(2/3)^6=320/729=0.43896\), or \(0.4390\) to four decimal places.
For \(F_A\), the far-right density is proportional to \(x^{-(2/2+1)}=x^{-2}\). For \(F_B\), it is proportional to \(x^{-(8/2+1)}=x^{-5}\). Thus \(F_A\) has the more slowly decaying far-right density tail. The values at \(x=1\) need not follow the eventual tail comparison: \(f_B(1)>f_A(1)\), even though \(F_B\)'s far-right tail decays faster.
Check. The modes are \(1/4\) and \(2/5\), the densities at 1 are \(0.2963\) and \(0.4390\), and the far-right powers are \(-2\) and \(-5\). The common error to avoid is treating a density comparison at one point as a comparison of the entire right tails.
How to Interpret Changes in Degrees of Freedom
The mode formula gives a useful summary of the two parameters’ combined effect. When \(\nu_1>2\), increasing the numerator degrees of freedom raises the factor \((\nu_1-2)/\nu_1\) toward 1. Increasing the denominator degrees of freedom raises \(\nu_2/(\nu_2+2)\) toward 1. For any finite degrees of freedom, both factors are less than 1, so the interior mode lies below 1.
The separate endpoint and tail calculations are often more informative than trying to describe every change with one word such as “less skewed.” The F density is asymmetric and has a right tail; its far-right density decays by a power, not by an exponential. A smaller denominator degree of freedom means that this power decays more slowly. A numerator degree of freedom at or below 2 creates a qualitatively different left endpoint from one above 2.
When identifying a shape from a graph or formula, first check that the support is positive. Next inspect the numerator degrees of freedom to classify the density at zero and determine whether the mode is interior. Finally inspect the denominator degrees of freedom to describe the far-right tail. This sequence keeps three distinct features from being confused.
Check Your Understanding
For each problem, identify the relevant degrees of freedom before describing the density.
- Let \(F\sim F_{2,6}\). State its support, its limiting density behavior at zero, and whether it has a positive interior mode. (Answer: Support \(x>0\); the density approaches a finite positive value at zero; there is no positive interior mode.)
- Find the mode of \(F\sim F_{8,6}\). (Answer: \(\frac{6}{8}\frac{6}{8}=\frac{9}{16}=0.5625\).)
- Compare the far-right density powers for \(F_{4,2}\) and \(F_{4,8}\). Which has the more slowly decaying density tail? (Answer: The powers are \(x^{-2}\) and \(x^{-5}\), respectively; \(F_{4,2}\) has the more slowly decaying tail.)
- For \(F\sim F_{3,5}\), determine the behavior at zero and calculate the mode. (Answer: The density approaches zero at zero; the mode is \(\frac{1}{3}\frac{5}{7}=\frac{5}{21}\approx0.2381\).)
- Compare \(F_{2,4}\) and \(F_{6,4}\) at the left endpoint. Which density approaches a finite positive value, and which approaches zero? (Answer: \(F_{2,4}\) approaches a finite positive value; \(F_{6,4}\) approaches zero.)