+ Reply to Thread
Results 1 to 8 of 8

Thread: Expected Value of a Sample Range

  1. #1
    Points: 3,039, Level: 33
    Level completed: 93%, Points required for next Level: 11

    Location
    Michigan, USA (GMT-5)
    Posts
    43
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Expected Value of a Sample Range




    The function for the expected value of a sample range is given on the following page.

    http://www.jmu.edu/docs/sasdoc/sasht...hapc/sect9.htm

    I'd be interested to know which distribution is used in the calculation of this expectation and how you get from the general formulation of an expectation (Integral x*f(x)dx, -%INF, %INF) to the final result.

    An original paper by Tippett(1925) is on JSTOR, but unfortuneately I have no access to it.

  2. #2
    Dark Knight
    Points: 2,328, Level: 29
    Level completed: 19%, Points required for next Level: 122
    vinux's Avatar
    Posts
    1,891
    Thanks
    45
    Thanked 193 Times in 160 Posts
    The sample range is the difference between the maximum and minimum.
    It is easy..

    sample range = max - min
    Let Fx is the CDF

    Distribution of Min is 1- (1- Fx)^n
    Distribution of Max is Fx^n

    Expected range = E(max) - E(min)
    In the long run, we're all dead.

  3. #3
    Dark Knight
    Points: 2,328, Level: 29
    Level completed: 19%, Points required for next Level: 122
    vinux's Avatar
    Posts
    1,891
    Thanks
    45
    Thanked 193 Times in 160 Posts
    There is one more formula for expectation .

    E(X) = integral ( [1- Fx] dx )

    This you can prove using integration by parts.


    Use this formula in the above you will get the expression for d2
    In the long run, we're all dead.

  4. #4
    Points: 3,039, Level: 33
    Level completed: 93%, Points required for next Level: 11

    Location
    Michigan, USA (GMT-5)
    Posts
    43
    Thanks
    0
    Thanked 0 Times in 0 Posts
    Quote Originally Posted by vinux View Post
    The sample range is the difference between the maximum and minimum.
    It is easy..

    sample range = max - min
    Let Fx is the CDF

    Distribution of Min is 1- (1- Fx)^n
    Distribution of Max is Fx^n

    Expected range = E(max) - E(min)
    I did know what a sample range was, but I'm curious about the distributions of min and max. Where did they come from? Also I think your signs are different than the ones in the link. The integrand there is
    Code: 
    1 - (1 - Fx)^n - (Fx)^n
    
    which is not equal to
    
    (Fx)^n - (1 - (1 - Fx)^n)
    As I mentioned in my original post an expectation should look like an integral from minus infinity to plus infinity of the independent variable x times the pdf of the distribution. In the formula you gave it is hard to see the explicit presence of x.
    Last edited by Papabravo; 10-15-2008 at 11:49 PM.

  5. #5
    Dark Knight
    Points: 2,328, Level: 29
    Level completed: 19%, Points required for next Level: 122
    vinux's Avatar
    Posts
    1,891
    Thanks
    45
    Thanked 193 Times in 160 Posts
    Quote Originally Posted by Papabravo View Post
    I did know what a sample range was, but I'm curious about the distributions of min and max. Where did they come from?
    The distribution part is not difficult.

    Let X,.X2..Xn are the iid(indpendent and identically distributed) r.v.s
    with CDF Fx
    let Ma is the maximum
    P[Ma <=x ]= P[X1<=x,.....Xn<=x] = product(i=1:n)(P[Xi<=x]) = Fx^n ( identical)

    For deriving Minimum, use the Reliability function
    Let Mi is the minimum
    P(Min>x) = P[X1>x,.....Xn>x]
    etc.

    See this link for some more information
    http://en.wikipedia.org/wiki/Order_statistic


    Quote Originally Posted by Papabravo View Post
    Also I think your signs are different than the ones in the link.



    As I mentioned in my original post an expectation should look like an integral from minus infinity to plus infinity of the independen variable x times the pdf of the distribution. In the formula you gave it is hard to see the explicit presence of x.
    I think you didn't read my second post.

    There is one more formula for expectation .
    E(X) = integral ( [1- Fx] dx )
    This you can prove using integration by parts.
    Use this formula in the above you will get the expression for d2
    In the long run, we're all dead.

  6. #6
    Points: 3,039, Level: 33
    Level completed: 93%, Points required for next Level: 11

    Location
    Michigan, USA (GMT-5)
    Posts
    43
    Thanks
    0
    Thanked 0 Times in 0 Posts
    Of course I saw your second post. The information on the order statistics was useful. Still I can't get from that formulation of the pdf for the minimum and the maximum to the expression in the original link. Here's the problem:
    In the orginal link the integrand involved only constants and powers of the normal CDF. The integrand you write for the expected value has the independant variable x, the pdf f(x), and powers of the CDF F(x) and the complement (1-F(x)). That is the remaining step that I'm missing. There is also the matter of the constant n! /(r-1)!*(n-r)! in the initial expression for the expectation where r = 1 for the minimum and r = n for the maximum.

  7. #7
    Dark Knight
    Points: 2,328, Level: 29
    Level completed: 19%, Points required for next Level: 122
    vinux's Avatar
    Posts
    1,891
    Thanks
    45
    Thanked 193 Times in 160 Posts
    Quote Originally Posted by Papabravo View Post
    Of course I saw your second post. The information on the order statistics was useful. Still I can't get from that formulation of the pdf for the minimum and the maximum to the expression in the original link. Here's the problem:
    In the orginal link the integrand involved only constants and powers of the normal CDF. The integrand you write for the expected value has the independant variable x, the pdf f(x), and powers of the CDF F(x) and the complement (1-F(x)). That is the remaining step that I'm missing. There is also the matter of the constant n! /(r-1)!*(n-r)! in the initial expression for the expectation where r = 1 for the minimum and r = n for the maximum.
    Let
    F1(x) is the CDF of Min &
    F2(x) is the CDF of Max

    E(Min)=(Integral ( [1-F1(x) ] dx, -%INF, %INF)
    E(Max)=(Integral ( [1-F2(x) ] dx, -%INF, %INF)

    E(Range) =E(max)-E(min)
    = (Integral ( [1-F2(x) -1 + F1(x) ] dx, -%INF, %INF)
    = (Integral ( [F1(x) - F2(x) ] dx, -%INF, %INF)

    I have already explaind the formulation of CDF of Min and Max

    Code: 
    let Ma is the maximum 
    P[Ma <=x ]= P[X1<=x,.....Xn<=x] = product(i=1:n)(P[Xi<=x]) = Fx^n ( identical)
    
    For deriving Minimum, use the Reliability function
    Let Mi is the minimum
    P(Min>x) = P[X1>x,.....Xn>x]
    etc.
    ie F1(x) = 1-{1 - F(x)}^n
    F2(x) = F(x)^n

    Replace this, you will get the answer.

    you have to offer a beer for more explanation.
    In the long run, we're all dead.

  8. #8
    Points: 3,039, Level: 33
    Level completed: 93%, Points required for next Level: 11

    Location
    Michigan, USA (GMT-5)
    Posts
    43
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Problem is I still don't believe that your espressions for E(Min) and E(Max) are correct. That being the case the rest of your argument is hand waving to fit the result. Thanks for trying, but I guess I'll have to dig out a copy of Tippett.

+ Reply to Thread

           




Similar Threads

  1. MLE of range of Uniform(a,b)
    By conundrum in forum Probability
    Replies: 9
    Last Post: 03-17-2011, 12:13 PM
  2. Replies: 6
    Last Post: 11-08-2010, 04:50 PM
  3. Interquartile range: Why add 1 to the sample size
    By FluffyStats in forum Statistics
    Replies: 1
    Last Post: 10-30-2010, 02:52 PM
  4. restriction of range
    By psychologygirl in forum Statistics
    Replies: 2
    Last Post: 06-01-2009, 04:03 PM
  5. Studentised range value
    By UK2008 in forum Statistics
    Replies: 4
    Last Post: 11-19-2008, 03:05 PM

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