My Math Forum  

Go Back   My Math Forum > College Math Forum > Advanced Statistics

Advanced Statistics Advanced Probability and Statistics Math Forum


Reply
 
LinkBack Thread Tools Display Modes
January 19th, 2010, 06:32 AM   #1
Newbie
 
Joined: May 2009

Posts: 25
Thanks: 0

Rule of succession problem

Taken from an example in the 1870 version of Todhunter algebra. I spotted and answered this problem (correctly?) on Yahoo answers....

You have a bag of 5 equally sized balls the colour of which you have no information about.
You draw two balls and both are white.

What is the probability that all five of the balls are white ?

I answered the problem without resorting to the succession rule.....how would you solve it ?
CarlPierce is offline  
 
January 19th, 2010, 07:11 AM   #2
Global Moderator
 
CRGreathouse's Avatar
 
Joined: Nov 2006
From: UTC -5

Posts: 16,046
Thanks: 938

Math Focus: Number theory, computational mathematics, combinatorics, FOM, symbolic logic, TCS, algorithms
Re: Rule of succession problem

I have no idea how to solve the problem. I suppose that any solution must infer a reasonable Bayesian prior...?

Link: Algebra for the Use of Colleges and Schools: With Numerous Examples, p. 466 ff.
CRGreathouse is offline  
January 19th, 2010, 10:36 AM   #3
Math Team
 
Joined: Dec 2006
From: Lexington, MA

Posts: 3,267
Thanks: 408

Re: Rule of succession problem

Hello, CarlPierce!

CRGreathouse is correct . . . This requires Bayes' Theorem.


Quote:
You have a bag of 5 equally sized balls the colour of which you have no information about.
You draw two balls and both are white.

What is the probability that all five of the balls are white ?

We know there are at least 2 Whites among the 5 balls.

I assume that the assortment colors of the balls are equally likely.










[color=beige]. . [/color]




[color=beige]. . [/color]

[color=beige]. . [/color]

[color=beige]. . [/color]

[color=beige]. . [/color]






soroban is offline  
January 19th, 2010, 12:51 PM   #4
Global Moderator
 
CRGreathouse's Avatar
 
Joined: Nov 2006
From: UTC -5

Posts: 16,046
Thanks: 938

Math Focus: Number theory, computational mathematics, combinatorics, FOM, symbolic logic, TCS, algorithms
Re: Rule of succession problem

Quote:
Originally Posted by soroban
This strikes me as a questionable assumption, though. Why not assume, say, that the prior probability that a ball is white is uniformly distributed on [0, 1]? That is, there is a universe with Bp white balls and B(1-p) nonwhite balls for B very large, and then the five balls are chosen with that probability based on your universe. This seems more natural to me, and probably gives a different result.
CRGreathouse is offline  
January 21st, 2010, 07:37 AM   #5
Newbie
 
Joined: May 2009

Posts: 25
Thanks: 0

Re: Rule of succession problem

I solved it as follows

Given 2 whites then the actual number of whites in the bag can be either 2,3,4 or 5

For each possible case the chance of drawing two whites is ...

2/20 , 6/20, 12/20 or 1

So the probability that the actual number of whites is 5 is therefore

1 / (2/20 + 6/20 + 12/20 + 1) = 0.5

The succession rule gives the same answer.
namely 3/4 x 4/5 x 5/6 = 0.5

Is my reasoning valid ? Todhunters gives the answer 0.5
CarlPierce is offline  
January 21st, 2010, 08:07 AM   #6
Global Moderator
 
CRGreathouse's Avatar
 
Joined: Nov 2006
From: UTC -5

Posts: 16,046
Thanks: 938

Math Focus: Number theory, computational mathematics, combinatorics, FOM, symbolic logic, TCS, algorithms
Re: Rule of succession problem

Quote:
Originally Posted by CarlPierce
Is my reasoning valid ?
Yes, if you assume that the prior odds of there being 2, 3, 4, or 5 white balls are equally likely. Todhunter and soroban seem to think that this is reasonable. I don't.

First, it seems to rest on the assumption that, a priori, a ball has even odds to be white or non-white. I question this.

But given that assumption, I would expect 1/32 prior odds that all 5 are white, not 1/4 or 1/6. This would give
1/(1*1 + 5*3/5 + 10*3/10 + 10*1/10) = 1/8
rather than 1/2.

Laplace's "rule of succession" would suggest that the chance that a random ball is white is (2 + 1) / (2 + 2) = 3/4, making the chance that the three unknown balls are white 3/4 * 3/4 * 3/4. The same analysis could of course be applied here:
# white balls (prior probability)
5 (3/4)^5 * (1/4)^0 * 1 = 243/1024
4 (3/4)^4 * (1/4)^1 * 5 = 405/1024
3 (3/4)^3 * (1/4)^2 * 10 = 270/1024
2 (3/4)^2 * (1/4)^3 * 10 = 90/1024
1 (3/4)^1 * (1/4)^4 * 5 = 15/1024
0 (3/4)^0 * (1/4)^5 * 1 = 1/1024

243/(243*1 + 405*3/5 + 270*3/10 + 90*1/10) = 27/64

The answers are the same either way.

I tend to think that 27/64 is more reasonable than 1/8, but I certainly wouldn't go any higher than 27/64. In particular I'm not willing to go up to the Todhunter 32/64.

The method of Agresti & Coull seems to suggest a yet lower figure: 8/27. The Clopper-Pearson 95% interval is maddeningly wide: 1.118% to 100%.
CRGreathouse is offline  
January 21st, 2010, 08:27 AM   #7
Newbie
 
Joined: May 2009

Posts: 25
Thanks: 0

Re: Rule of succession problem

I'm not sure I need to assume anything about the prior probabilities of a ball being white.


I simple state that if 2 white balls have been drawn then the respective changes it happened given the four possible states were

2/20, 6/20, 12/20 or 1

Can't we turn the odds around to infer the chance of each state actually being true.

So is it not reasonable to infer that the P(white = 5) is the ratio of these ? i.e 20/(2+6+12+20) = 0.5
CarlPierce is offline  
January 21st, 2010, 08:36 AM   #8
Global Moderator
 
CRGreathouse's Avatar
 
Joined: Nov 2006
From: UTC -5

Posts: 16,046
Thanks: 938

Math Focus: Number theory, computational mathematics, combinatorics, FOM, symbolic logic, TCS, algorithms
Re: Rule of succession problem

Quote:
Originally Posted by CarlPierce
I'm not sure I need to assume anything about the prior probabilities of a ball being white.
You do.

Quote:
Originally Posted by CarlPierce
I simple state that if 2 white balls have been drawn then the respective changes it happened given the four possible states were

2/20, 6/20, 12/20 or 1

Can't we turn the odds around to infer the chance of each state actually being true.
Suppose that you happened to know that someone (say, the Easter bunny; alternatively, Descartes' demon) went around the world leaving bags of five white balls everywhere. In fact the bunny leaves so many bags around that the chance you'd come across another bag is infinitesimal. In that case the probabilities would still be
2/20, 6/20, 12/20 or 1
as you describe, but the only reasonable answer would be 1.

Suppose instead that you knew the bunny left three kinds of bags: one with 5 white, one with 3 white, and one with 1 white; and that the respective probabilities were 1/4, 1/2, and 1/4. Then you would be justified in deriving a probability of 5/8.

Quote:
Originally Posted by CarlPierce
So is it not reasonable to infer that the P(white = 5) is the ratio of these ? i.e 20/(2+6+12+20) = 0.5
Only if the four possibilities were equiprobable.
CRGreathouse is offline  
January 21st, 2010, 09:40 AM   #9
Senior Member
 
Joined: Feb 2009
From: Adelaide, Australia

Posts: 1,519
Thanks: 3

Re: Rule of succession problem

Quote:
You have a bag of 5 equally sized balls the colour of which you have no information about.
No information is no information.

Yesterday was a weekday, and today is a weekday. What is the probability the next three days will be weekdays?
aswoods is offline  
January 21st, 2010, 12:12 PM   #10
Global Moderator
 
CRGreathouse's Avatar
 
Joined: Nov 2006
From: UTC -5

Posts: 16,046
Thanks: 938

Math Focus: Number theory, computational mathematics, combinatorics, FOM, symbolic logic, TCS, algorithms
Re: Rule of succession problem

Quote:
Originally Posted by aswoods
No information is no information.

Yesterday was a weekday, and today is a weekday. What is the probability the next three days will be weekdays?
The so-called "Sunrise Problem".
CRGreathouse is offline  
Reply

  My Math Forum > College Math Forum > Advanced Statistics

Tags
problem, rule, succession



Search tags for this page
Click on a term to search for related topics.
Thread Tools
Display Modes


Similar Threads
Thread Thread Starter Forum Replies Last Post
chain rule problem geryuu Calculus 2 December 21st, 2013 07:14 PM
Succession four cubic powers mente oscura Number Theory 4 June 4th, 2013 11:13 AM
General succession for Pythagorean numbers mente oscura Number Theory 4 May 30th, 2013 04:27 PM
The Product Rule Problem mathkid Calculus 32 November 21st, 2012 02:41 PM
L'Hopital's Rule (Second Problem!) David_Lete Calculus 1 January 11th, 2011 01:34 AM





Copyright © 2019 My Math Forum. All rights reserved.