+ Reply to Thread
Results 1 to 5 of 5

Thread: Multivariate distributions that are 'closed' under marginalization?

  1. #1
    TS Contributor
    Points: 22,395, Level: 93
    Level completed: 5%, Points required for next Level: 955
    spunky's Avatar
    Location
    vancouver, canada
    Posts
    2,135
    Thanks
    166
    Thanked 537 Times in 431 Posts

    Multivariate distributions that are 'closed' under marginalization?




    a very, very nice property of the multivariate normal distribution is that any lower-dimensional margins are also multivariate (or univariate, depending on how low you go) normally distributed.

    does anyone know any other distributions for which this property is true?

    is there some sort of 'general rule' for this? like if a multivariate distribution satisfies such and such properties, then its lower-dimensional marginals are also the same distirbution (but in lower dimensions)?
    for all your psychometric needs! https://psychometroscar.wordpress.com/about/

  2. #2
    Points: 2,109, Level: 27
    Level completed: 73%, Points required for next Level: 41
    ted00's Avatar
    Location
    USA
    Posts
    237
    Thanks
    21
    Thanked 29 Times in 25 Posts

    Re: Multivariate distributions that are 'closed' under marginalization?

    Here's a bivariate Poisson whose marginals are Poisson
    http://en.wikipedia.org/wiki/Poisson...n_distribution

    From what I remember, there's not just "one" bivariate Poisson, but several people have developed several different bivariate (and trivariate) distributions that have the appropriate properties (I think some of them are listed in the "See Other" of that wikipedia page)

    Multivariate probability distributions are a weird thing; I think the Poisson is a good example -- there's not always a clear "multivariate version" of each univariate distribution, but people can develop multivariate distributions that have certain properties that resemble a univariate distribution, but these may not be unique (as in the multivariate Poisson cases)
    The mathematical explanation of a statistical procedure is really just pseudo-code, which we can make operational by translating it into real computer code. --B. Klemens

  3. #3
    TS Contributor
    Points: 22,395, Level: 93
    Level completed: 5%, Points required for next Level: 955
    spunky's Avatar
    Location
    vancouver, canada
    Posts
    2,135
    Thanks
    166
    Thanked 537 Times in 431 Posts

    Re: Multivariate distributions that are 'closed' under marginalization?

    Dason also suggested the multivariate Bernoulli distribution... but after some exhaustive Google searches it seems that the only way to find this out is integrate some dimensions out and hope that whatever results can be expressed as the family of distributions one is interested in.

    i was hoping there would be some obscure, hidden theorem out there or some quick check you could do on the PDF... but there aren't none...

    NONE!!!
    for all your psychometric needs! https://psychometroscar.wordpress.com/about/

  4. #4
    TS Contributor
    Points: 22,395, Level: 93
    Level completed: 5%, Points required for next Level: 955
    spunky's Avatar
    Location
    vancouver, canada
    Posts
    2,135
    Thanks
    166
    Thanked 537 Times in 431 Posts

    Re: Multivariate distributions that are 'closed' under marginalization?

    well, apparently under a specific partition the multivariate t distribution has also t-distributed marginals
    for all your psychometric needs! https://psychometroscar.wordpress.com/about/

  5. #5
    Points: 2,109, Level: 27
    Level completed: 73%, Points required for next Level: 41
    ted00's Avatar
    Location
    USA
    Posts
    237
    Thanks
    21
    Thanked 29 Times in 25 Posts

    Re: Multivariate distributions that are 'closed' under marginalization?


    Yes, Kotz, that a name I was trying to remember

    There's a series of books by Kotz, N. Balakrishnan, and Norman L. Johnson on Univariate and Multivariate continuous and discrete distributions that might be of interest to you; since you're a grad student (if that's up to date) should be no problem getting your hands on them. Lots of info in them you might be interested in, including how to derive obscure distributions
    Last edited by ted00; 03-15-2014 at 11:01 PM.
    The mathematical explanation of a statistical procedure is really just pseudo-code, which we can make operational by translating it into real computer code. --B. Klemens

+ Reply to Thread

           




Tags for this 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