+ Reply to Thread
Results 1 to 5 of 5

Thread: Analytic form for the quadrant probability of a (standard) bivariate t distribution?

  1. #1
    TS Contributor
    Points: 22,432, Level: 93
    Level completed: 9%, Points required for next Level: 918
    spunky's Avatar
    Location
    vancouver, canada
    Posts
    2,135
    Thanks
    166
    Thanked 537 Times in 431 Posts

    Analytic form for the quadrant probability of a (standard) bivariate t distribution?




    hellows! i may need this further along my manuscript and quick google searches aren't really being helpful, so maybe someone around here knows.

    for the case of the bivariate standard normal distribution, the quadrant probability is given by:

    P(x_{1}\leq0,x_{2}\leq0)=P(x_{1}\geq0,x_{2}\geq0)=\int_{-\infty }^{0}\int_{-\infty }^{0}P(x_{1},x_{2})dx_{1}dx_{2}=\frac{1}{4}+\frac{sin^{-1}\rho}{2\pi}

    (or at least the quadrants going in the same direction).

    i may find myself needing a nice, close-formed thingy like that one but for the (standard) bivariate t distribution, but goolging stuff like that doesn't give me many results that i could use.

    does anyone know if there is a nice closed formed expression for the (standard) bivariate t distribution like there is one for the bivariate standard normal distribution? or there isn't any?

    thanks!
    for all your psychometric needs! https://psychometroscar.wordpress.com/about/

  2. The Following User Says Thank You to spunky For This Useful Post:

    GretaGarbo (03-02-2016)

  3. #2
    Cookie Scientist
    Points: 13,431, Level: 75
    Level completed: 46%, Points required for next Level: 219
    Jake's Avatar
    Location
    Austin, TX
    Posts
    1,293
    Thanks
    66
    Thanked 584 Times in 438 Posts

    Re: Analytic for the quadrant probability of a (standard) bivariate t distribution?

    It seems unlikely since, according to wiki, there is no analytic expression for the CDF of multivariate t.
    “In God we trust. All others must bring data.”
    ~W. Edwards Deming

  4. #3
    TS Contributor
    Points: 22,432, Level: 93
    Level completed: 9%, Points required for next Level: 918
    spunky's Avatar
    Location
    vancouver, canada
    Posts
    2,135
    Thanks
    166
    Thanked 537 Times in 431 Posts

    Re: Analytic for the quadrant probability of a (standard) bivariate t distribution?

    Quote Originally Posted by Jake View Post
    It seems unlikely since, according to wiki, there is no analytic expression for the CDF of multivariate t.
    but... there is also no analytic expression for the CDF of the multivariate normal distribution. yet there is a closed-form version for that specific probability quadrant.

    there must be something i'm missing....
    for all your psychometric needs! https://psychometroscar.wordpress.com/about/

  5. #4
    TS Contributor
    Points: 12,227, Level: 72
    Level completed: 45%, Points required for next Level: 223
    rogojel's Avatar
    Location
    I work in Europe, live in Hungary
    Posts
    1,470
    Thanks
    160
    Thanked 332 Times in 312 Posts

    Re: Analytic for the quadrant probability of a (standard) bivariate t distribution?

    hi,
    on an off chance - we used to use a book by the authors Abramovici and Stegun (something like a hanbook of series and integrals) way before google was implemented It used to be, that if a formula is not in the A/S it does not exist.

  6. The Following User Says Thank You to rogojel For This Useful Post:

    GretaGarbo (03-02-2016)

  7. #5
    TS Contributor
    Points: 22,410, Level: 93
    Level completed: 6%, Points required for next Level: 940

    Posts
    3,020
    Thanks
    12
    Thanked 565 Times in 537 Posts

    Re: Analytic form for the quadrant probability of a (standard) bivariate t distributi


    I think most of you already know this but let's present this once again before trying to generalize further.

    Let Z_1, Z_2 be i.i.d. standard normal random variables. From Cholesky Decomposition, we know that

    X_1 = Z_1, X_2 = \rho Z_1 + \sqrt{1 - \rho^2} Z_2

    will jointly follows a bivariate standard normal with correlation \rho \in (-1, 1). The joint pdf will be in terms of

    \begin{bmatrix} x_1 & x_2 \end{bmatrix} 
\begin{bmatrix} 1 & \rho \\ \rho & 1 \end{bmatrix}^{-1} 
\begin{bmatrix} x_1 \\ x_2 \end{bmatrix}


    (*)
    Now for any (X_1, X_2) with joint pdf solely depending the above quadratic form, we try to use the same back-trasform listed in above into (Z_1, Z_2),
    and see how it affect the joint pdf:

    \frac {\partial x_1} {\partial z_1} = 1,
\frac {\partial x_1} {\partial z_2} = 0,
\frac {\partial x_2} {\partial z_1} = \rho,
\frac {\partial x_2} {\partial z_2} = \sqrt{1 - \rho^2}

    and thus the Jacobian is
    |1 \times \sqrt{1 - \rho^2} - 0 \times \rho| = \sqrt{1 - \rho^2}

    and independent of z_1, z_2. The resulting joint pdf will be in terms of

    \begin{bmatrix} z_1 & \rho z_1 + \sqrt{1 - \rho^2}z_2 \end{bmatrix}
\begin{bmatrix} 1 & \rho \\ \rho & 1 \end{bmatrix}^{-1} 
\begin{bmatrix} z_1 \\ \rho z_1 + \sqrt{1 - \rho^2}z_2 \end{bmatrix}

    = \frac {1} {1 - \rho^2}\begin{bmatrix} z_1 & z_2 \end{bmatrix} 
\begin{bmatrix} 1 & \rho \\ 0 & \sqrt{1 - \rho^2} \end{bmatrix}
\begin{bmatrix} 1 & -\rho \\ -\rho & 1 \end{bmatrix}
\begin{bmatrix} 1 & 0 \\ \rho & \sqrt{1 - \rho^2} \end{bmatrix}
\begin{bmatrix} z_1 \\ z_2 \end{bmatrix}

    = \frac {1} {1 - \rho^2}\begin{bmatrix} z_1 & z_2 \end{bmatrix} 
\begin{bmatrix} 1 - \rho^2 & 0 \\ 0 & 1 - \rho^2 \end{bmatrix}
\begin{bmatrix} z_1 \\ z_2 \end{bmatrix}

    = \begin{bmatrix} z_1 & z_2 \end{bmatrix} 
\begin{bmatrix} z_1 \\ z_2 \end{bmatrix}

    = z_1^2 + z_2^2

    So by the transformation, we see that the joint pdf of (Z_1, Z_2) has a circular contour, and thus is rotational symmetric about the mean (i.e. the origin). Therefore for any pair of straight lines intersecting at the mean and dividing \mathbb{R}^2 into four regions, the probability that (Z_1, Z_2) fall into a particular region will be equal to the proportion of the angle at centre from 2\pi, similar to the case when we calculating the area of a sector from circle.

    And that is why we have the very neat closed-form result:

    \Pr\{X_1 > 0, X_2 > 0\}

    = \Pr\{Z_1 > 0, \rho Z_1 + \sqrt{1 - \rho^2} Z_2 > 0\}

    = \Pr\left\{Z_1 > 0, Z_2 > - \frac {\rho} {\sqrt{1 - \rho^2}} Z_1 \right\}

    Now the region enclosed by

    \left\{z_1 > 0, z_2 > - \frac {\rho} {\sqrt{1 - \rho^2}} z_1 \right\}

    in the (z_1, z_2)-plane consist of two bounds, one is the vertical z_2-axis, and the other one is a straight line passing through origin with slope - \frac {\rho} {\sqrt{1 - \rho^2}}. The angle at centre will be

    \frac {\pi} {2} - \tan^{-1}\left(- \frac {\rho} {\sqrt{1 - \rho^2}}\right) 
= \frac {\pi} {2} + \sin^{-1}\rho

    Dividing by 2\pi yields Spunky posted result.


    Now back to the question: In the above calculation starting from (*), we merely required that the joint pdf solely in terms of the quadratic form (we do not need to require it to be multivariate normal), so it should holds for any ellipitical distribution, including the standard bivariate t-distribution mentioned. The bivariate t-distribution mentioned was referring to
    https://en.wikipedia.org/wiki/Multiv...ion#Derivation

    So my guess will be the same result also applies to all elliptical distributions (note that in bivariate t, the joint pdf has a circular contour does not mean Z_1, Z_2 are independent). I hope this guess is correct Feel free to point out if the argument is wrong.

  8. The Following User Says Thank You to BGM For This Useful Post:

    GretaGarbo (03-02-2016)

+ Reply to Thread

           




Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts






Advertise on Talk Stats