home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky sci.math:18854 sci.math.stat:2891
- Path: sparky!uunet!charon.amdahl.com!pacbell.com!sgiblab!swrinde!dptspd!tamsun.tamu.edu!ksanjay
- From: ksanjay@cs.tamu.edu (Sanjay D Kamat)
- Newsgroups: sci.math,sci.math.stat
- Subject: Request for help - A Probability Problem
- Date: 27 Jan 1993 19:17:37 GMT
- Organization: Computer Science Department, Texas A&M University
- Lines: 42
- Sender: Sanjay Kamat
- Distribution: World
- Message-ID: <1k6n4hINNd7@tamsun.tamu.edu>
- NNTP-Posting-Host: sparc29.cs.tamu.edu
- Summary: Help requested on a random sums problem
- Keywords: Probability, Beta distribution, Random Sums
-
-
- Hi,
- I am a computer science student working towards my PhD degree. I
- would very much appreciate your help/suggestions on the following
- problem I recently encountered in my research.
-
- A random variable S is defined as the sum of n terms, each of the
- form u_i * w_i.
- u_i and w_i are themselves random with the following properties.
-
- 1) All u_i's are identically distributed -but not independent in the
- sense that their sum is fixed, say U. (All splittings of U
- considerted equi-probable).
- 2) All w_i's are independent as well as identically distributed.
- They are also bounded between w_min and w_max.
- 3) u_i's and w_i's are independent of each other.
-
- I can establish that
- mean S = mean w * U
- variance S = (2/(n+1)) * U * U * variance w.
-
- Further, for the parameters of concern, "beta" distribution for S
- with above mean and variance seems to fit very well.
-
- My questions are:
-
- 1) Is there a sound theoretical justification for "beta"
- distribution to be applicable?
- ( Also - is there some limit theorem which applies to S for large
- n? )
- 2) What is a good source for good approximations for the incomplete
- beta function?
-
- I would appreciate if you email your reply to me.
-
- Thanks in advance,
-
- Sanjay Kamat.
- Dept of Computer Science
- Texas A&M University
- (ksanjay@cs.tamu.edu)
-
-