How to Derive the F Distribution

Math Proof, Statistics / By /
f-distribution from chi-squared distribution

Table of Contents


Context


F Distribution Definition

The F Distribution is the ratio of two independent Chi-squared distributions standardized by their degrees of freedom.
\[\large{
\begin{align}
Y_k\sim\chi^2(k) && X\sim F(p,q) = \frac{Y_p/p}{Y_q/q}
\end{align}
}\]


Single Variable Change

Similar to the T Distribution, deriving the F Distribution involves several single variable changes.

A variable change is a change to a functions variable and domain such that the new and old functions have integrals which are equal when evaluated over their respective domains (eg if you perform a variable change on a probability function you would still expect the area under the probability curve to be one).
\[\large{
\begin{align}
y = g(x) && \int_y f(y) \partial y = \int_x f(g(x)) \frac{\partial g}{\partial x}\partial x
\end{align}
}\]


Multi-variable Change with Jacobian Matrix

This proof involves one multi-variable change. In order to accomplish this the determinant of a Jacobian matrix is required.

**NOTE**: The combination of the Jacobian matrix and its determinant are often referred to together as simple the Jacobian.

A function with more than one scalar input variable can also be expressed as a function with a single vector input variable.
\[\large{
f(y_1, y_2, y_3) = f\Big(\stackrel{y}{\begin{bmatrix} y_1 \\ y_2 \\ y_3 \end{bmatrix}}\Big) = f(y)
}\]
A multi-variable change requires that the input vector for the original function can itself be expressed as the range of a vector valued function of the variable you wish to change to.
\[\large{
\begin{align}
\stackrel{y}{\begin{bmatrix}y_1\\ y_2 \end{bmatrix}} = g\Big(\stackrel{x}{\begin{bmatrix} x_1 \\ x_2 \\ x_3 \end{bmatrix}}\Big) = \stackrel{g(x)}{\begin{bmatrix} x_1 + x_2 + x_3 \\ 5x_1 x_3 \end{bmatrix}}
\end{align}
}\]
A Jacobian matrix is a matrix of the first order partial derivatives of a vector valued function with respect to its vector input domain.
\[\large{
\stackrel{g(x)}{\begin{bmatrix} g_1(x_1,\ldots,x_n) \\ \vdots \\ g_k(x_1,\ldots,x_n) \end{bmatrix}}\longrightarrow J_{g(x)} = \begin{bmatrix}
\frac{\partial g_1}{\partial x_1} & \ldots & \frac{\partial g_1}{\partial x_n} \\
\vdots & \ddots & \vdots \\
\frac{\partial g_k}{\partial x_1} & \ldots & \frac{\partial g_k}{\partial x_n}
\end{bmatrix}
}\]
If the determinant of the Jacobian matrix exists it can be used to perform a multi-variable change on functions where the input vector is the domain of the function the Jacobian matrix was built from.
\[\large{
\begin{align}
y = g(x) && \int_y f(y) \partial y = \int_x f(g(x))\cdot \text{det}(J_{g(x)}) \partial x && \frac{\partial f}{\partial y} = \frac{\partial f}{\partial g} \text{det}(J_{g(x)})
\end{align}
}\]


How to Derive the F Distribution

An F random variable is defined as the ratio of two Chi-Squared random variables standardized by their own degrees of freedom.

Let \(X\) be the ratio of two independent Chi-squared random variables, \(Y_p\) and \(Y_q\), standardized by their degrees of freedom parameters \(p\) and \(q\) respectively. Then \(X\) is an F random variable with density function for \(f_X(x;p,q) = \frac{(x)^{p/2-1}p^{p/2}q^{q/2}}{\text{B}(p/2,q/2)(xp+q)^{p/2+q/2}}\).

First declare two related equalities (because they share a variable).
\[\large{
\begin{align}
&W_1 = Y_p/Y_q& &\longrightarrow& &Y_p = W_1 W_2 && \normalsize (1) \\ \\
&W_2 = Y_q& &\longrightarrow& &Y_q = W_2 \\ \\
\end{align}
}\]
The above equalities can be expressed in vector form as below and vector y can be expressed as a function of vector w.
\[\large{
\begin{align}
\stackrel{w}{\begin{bmatrix} W_1 \\ W_2\end{bmatrix}} && \stackrel{y}{\begin{bmatrix} Y_p \\ Y_q \end{bmatrix}} = \stackrel{g(w)}{\begin{bmatrix} W_1W_2 \\ W_2 \end{bmatrix}} && \normalsize (2) \\ \\
\end{align}
}\]
\(Y_p\) and \(Y_q\) are independent random variables so their joint probability function is simply the product of their individual Chi-squared probability functions.
\[\large{
\begin{align}
f_{Y_p,Y_q}(y_p,y_q) &= \frac{y_p^{p/2-1}e^{-y_p/2}}{\Gamma(p/2)2^{p/2}}\cdot\frac{y_q^{q/2-1}e^{-y_q/2}}{\Gamma(q/2)2^{q/2}} && \normalsize (3) \\ \\
&= \frac{y_p^{p/2-1}y_q^{q/2-1}e^{-\frac{(y_p + y_q)}{2}}}{\Gamma(p/2)\Gamma(q/2)2^{p/2+q/2}}
\end{align}
}\]
The next step is to perform a multi-variable change on the joint probability function from \(y\) to \(w\) using the Jacobian (instructions above).
\[\large{
\begin{align}
&J_{g(w)} = \begin{bmatrix} \frac{\partial g_1}{\partial w_1} & \frac{\partial g_1}{\partial w_2} \\
\frac{\partial g_2}{\partial w_1} & \frac{\partial g_2}{\partial w_2}
\end{bmatrix} =\begin{bmatrix} w_2 & w_1 \\
0 & 1
\end{bmatrix} && \normalsize (4)\\ \\
&\text{det}(J_{g(w)}) = w_2
\end{align}
}\]
Express the integral of the joint probability function \(Y_p\) and \(Y_q\) as a function of vector \(y\), then perform the variable change from \(y\) to \(w\) using the Jacobian.
\[\require{\cancel}\large{
\begin{align}
f(y) &= f(g(w))\cdot\text{det}(J_{g(w)}) &&\normalsize (5) \\ \\

&= \frac{(w_1w_2)^{p/2-1}w_2^{q/2\cancel{-1}}e^{-\frac{(w_1w_2 + w_2)}{2}}}{\Gamma(p/2)\Gamma(q/2)2^{p/2+q/2}}\cdot \cancel{w_2} \\ \\

f_{W_1,W_2}(w_1,w_2;p,q)&= \frac{w_1^{p/2-1}w_2^{p/2+q/2-1}e^{-\frac{(w_1w_2 + w_2)}{2}}}{\Gamma(p/2)\Gamma(q/2)2^{p/2+q/2}}\partial w
\end{align}
}\]
Now integrate out random variable \(W_2\) to obtain the univariate probability function of \(W_1\).

In order to do this notice that now there is a potential gamma function kernel given by \(w_2^{p/2+q/2-1}e^{-w_2\frac{w_1+1}{2}}\). In order to complete the expression of the gamma function two steps are necessary. First insert multiply the function by 1 and factor it: \(\require{\color}{\color{aqua}1 = \Big(\frac{2(w_1+1)}{2(w_1+1)}\Big)^{p/2+q/2-1}}\).
\[\large{
\begin{align}
\int_0^\infty f_{W_1,W_2}(w_2;p,q)\partial w_2 &= \int_0^\infty \frac{w_1^{p/2-1}w_2^{p/2+q/2-1}e^{-w_2\frac{w_1 + 1}{2}}}{\Gamma(p/2)\Gamma(q/2)2^{p/2+q/2}}\cdot{\color{aqua}\Big(\frac{2(w_1+1)}{2(w_1+1)}\Big)^{p/2+q/2-1}} \partial w_2 &&\normalsize (6) \\ \\

&= \int_0^\infty \frac{w_1^{p/2-1}(w_2{\color{aqua}\frac{w_1+1}{2}})^{p/2+q/2-1}e^{-w_2\frac{w_1 + 1}{2}}}{\Gamma(p/2)\Gamma(q/2)2^{p/2+q/2}}\cdot{\color{aqua}\Big(\frac{2}{w_1+1}\Big)^{p/2+q/2-1}} \partial w_2
\end{align}
}\]
Step two to completing the gamma function expression is to perform another variable change, this time from \(W_2\) to \(t\). Let \(t = w_2\frac{w_1+1}{2}\), which implies \(W_2 = g(t) = \frac{2t}{w_1+1}\).
\[\large{
\begin{align}
\int_0^\infty f_{W_1,W_2}(w_2) \partial w_2 &= \int_0^\infty f_{W_1,W_2}(g(t))\cdot\frac{\partial g}{\partial t} \partial t &&\normalsize (7) \\ \\

&= {\color{gold}\int_0^\infty} \frac{w_1^{p/2-1}{\color{gold}t^{p/2+q/2-1}e^{-t}}}{\Gamma(p/2)\Gamma(q/2)2^{p/2+q/2}}\cdot\Big(\frac{2}{w_1+1}\Big)^{p/2+q/2\cancel{-1}}\cdot\cancel{\frac{2}{w_1+1}}{\color{gold}\partial t} \\ \\

&= \frac{w_1^{p/2-1}{\color{gold}\Gamma(p/2+q/2)}}{\Gamma(p/2)\Gamma(q/2)\cancel{2^{p/2+q/2}}}\cdot\Big(\frac{\cancel{2}}{w_1+1}\Big)^{p/2+q/2}
\end{align}
}\]
After obtaining \(\Gamma(p/2+q/2)\) it can be combined with the other two gamma functions into a beta function.
\[\large{
\begin{align}
f_{W_1}(w_1;p,q) &= \frac{w_1^{p/2-1}}{\text{B}(p/2,q/2)}\cdot\Big(\frac{1}{w_1+1}\Big)^{p/2+q/2} && \normalsize (8)
\end{align}
}\]
Recall from (1) that \(W_1\) is the unstandardized ratio random variable of two Chi-squared random variables. Let \(X = W_1\frac{q}{p}\), which makes it the standardized ratio of two Chi-squared random variables. Perform a final variable change from \(W_1\) to \(X\).
\[\large{
\begin{align}
f_{W_1}(w_1;p,q) &= f_{W_1}(g(x);p,q)\cdot\frac{\partial g}{\partial x} &&\normalsize(9) \\ \\
f_X(x;p,q) &= \frac{x^{p/2-1}(\frac{p}{q})^{p/2\cancel{-1}}}{\text{B}(p/2,q/2)}\cdot\Big(\frac{1}{x\frac{p}{q}+1}\Big)^{p/2+q/2}\cdot \cancel{\frac{p}{q}} \\ \\
&= \frac{x^{p/2-1}(\frac{p}{\cancel{q}})^{p/2}}{\text{B}(p/2,q/2)}\cdot\frac{q^{\cancel{p/2}+q/2}}{(xp+q)^{p/2+q/2}}\\ \\
&= \frac{x^{p/2-1}p^{p/2}q^{q/2}}{\text{B}(p/2,q/2)(xp+q)^{p/2+q/2}} &&\normalsize (\text{QED})
\end{align}
}\]


Summary Conclusion

Deriving the F distribution probability function from the Chi-squared distribution isn’t a complicated task, but it can get messy with 3 variable changes.

Start from the definition of an F random variable defined as the ratio of two independent Chi-squared random variables scaled by their degrees of freedom. Let the two Chi-squared be \(Y_p\) and \(Y_q\). Define two more random variables via two equivalencies \(W_1 = \frac{Y_p}{Y_q}Y_p = Y_p\quad W_2=Y_q\).

Use the equivalencies between the y and w random to perform a multi-variable change on the joint probability function \(f_Y(y)\). Next integrate out \(W_2\) by completing a gamma function kernel. This leaves the single variable probability function for \(W_1\).

\(W_1\) is defined as the unstandardized ratio random variable of two independent Chi-squared random variables. Perform a final variable change, simply applying the standardizing scalar to the random variable, to arrive at the probability function for an F random variable.

Related Posts

135 thoughts on “How to Derive the F Distribution”

  9. えろ 人形

    comのドールコレクションは非常に多様で、さまざまなニーズに対応しています.リアルなコンパニオンからユニークなアートピースまで、幅広い選択肢が揃っています.セックス ドール

  17. ダッチワイフ

    To briefly sum up Wikipedia’s page of this most provocative topic: “The buttocks have been considered an erogenous zone in Western thought for centuries,and the eroticization of the female buttocks was heteronormative and due to their association and closeness to the female reproductive organs [even though in another sense] the buttocks are often taboo due to their proximity to the anus and association with the excretory system.ラブドール 中古

  29. ラブドール

    but skilled gaslighters also avoid taking responsibility for more everyday behavior—though the newest research suggests that most of their avoidance tactics center around taking attention away from their immature,hurtful behaviors.ダッチワイフ

  32. ラブドール

    結論として、comは高品質でリアルなドールを購入するための完璧な場所です.中国 えろその卓越した品質、豊富なカスタマイズオプション、優れたカスタマーサービスが、新規の購入者や経験豊富なコレクターにとって最高の選択肢です.

  40. ラブドール

    and difficulties forming adult relationships.オナドールIt is also believed by many experts that clergy abuse can damage the faith lives of its victims; many survivors interviewed described their experiences as “soul murder.

  72. ラブドール

    プレミアムな機能を満載した豪華なドールをお探しでも、ラブドール エロ満足を提供する予算に優しい選択肢をお探しでも、ご安心ください。お客様のニーズはすべてカバーされています。

  91. ラブドール中古

    Below is my reasoned response (which went unanswered):Do you have any empirical evidence for the efficacy of your negative conditioning and aversive techniques for fetishes? Do you have supporting documentation on what percentage of fetishes “prevent,good,ダッチワイフ エロ

  96. sootoca

    precio priligy 30 mg com 20 E2 AD 90 20Viagra 20Tablet 20Delhi 20 20Ile 20Kosztuje 20Viagra 20Cena 20W 20Aptekach ile kosztuje viagra cena w aptekach If sustained, this may ultimately curb their demand trendor, in countries where oil subsidies are in place, raisepressure on their governments to reduce those subsidyprogrammes, the IEA said

  102. ラブドール えろ

    sex can sometimes unconsciously be engaged in to defend against love and intimacy.Someone who has been severely wounded during childhood in the way Guggenheim reportedly was would typically avoid situations in which they could be rejected and abandoned again.オナホ 高級

  108. リアルラブドール

    Nominal Threats: As good as intercourse with a human husband or wife might be, エロ 人形there is always the potential risk of contracting sexually transmitted conditions or accidentally obtaining Expecting, not forgetting the risks connected to promiscuous sexual conduct with random associates or sex staff.

  111. オナホ おすすめ

    直感的なインターフェースにより、目的のドールを簡単に見つけることができるため、ストレスフリーなショッピング体験を提供します.セックス ドール各製品のページには、詳細な商品説明と高品質な画像が豊富に掲載されており、購入前にすべての情報を確認できます.

  112. えろ

    No interaction between the sexes is necessary for intrasexual conflict to occur.中国 エロInterlocus conflict happens when the interaction between males and females benefits one while is costly or detrimental to the other.

  115. エロ

    オナホ, including giving and getting consent for sexual behaviors)Children and adolescents need accurate information, critical skills, and relevant supports to ensure healthy and appropriate sexual behavior on their part.Preventing Child Sexual Abuse: What Parents Need to Know The idea that any adult would sexually abuse a child is terrifying,

Leave a Comment

Your email address will not be published. Required fields are marked *