3 prisoner puzzle

Another way of looking at it is.
For your $1 bet there are 4 outcomes. You lose, you win $1, you win $2 and you win $3.

The odds of you losing are 125/216
5/6 x 5/6 x 5/6 = 125/216

The odds of winning $1 are 75/216

1/6 x 5/6 x 5/6 = 25/216 +
5/6 x 1/6 x 5/6 = 25/216 +
5/6 x 5/6 x 1/6 = 25/216

The odds of winning $2 are 15/216

1/6 x 1/6 x 5/6 = 5/216 +
1/6 x 5/6 x 1/6 = 5/216 +
5/6 x 1/6 x 1/6 = 5/216

The odds of winning £3 are 1/216

1/6 x 1/6 x 1/6 = 1/216

The overall winnings are therefore

(75/216 x $1) + (15/216 x $2) + (1/216 x $3) =

$75/216 + $30/216 + $3/216 = £108/216 = 50 cents.

The house will on average pay out half the stakes and keep half.
 
Oh no, I've been completely misunderstood.

Sorry, folks. Let me try again.

If say, you bet on the number 1. And the 3 die show 2, 3 and 4, you lose your bet.

If say, you bet on the numbers 1, 2 and 3 and the 3 die show 4, 5 and 6, you lose all 3 bets.

If say, you bet on the numbers 1, 2 and 3 and the 3 die show 1, 5 and 6, you win the bet on number 1 but lose the other 2.

To all appearances, it looks like a 50-50 thing between the house and the punter. 3 die, 6 numbers on the board - doesn't it look like 50-50?

Forgive me, what I didn't take into account is that this game is completely alien to most of you. Here in Singapore, most would immediately know what I'm talking about.
 
Another way of looking at it is.
For your $1 bet there are 4 outcomes. You lose, you win $1, you win $2 and you win $3.

The odds of you losing are 125/216
5/6 x 5/6 x 5/6 = 125/216

The odds of winning $1 are 75/216

1/6 x 5/6 x 5/6 = 25/216 +
5/6 x 1/6 x 5/6 = 25/216 +
5/6 x 5/6 x 1/6 = 25/216

The odds of winning $2 are 15/216

1/6 x 1/6 x 5/6 = 5/216 +
1/6 x 5/6 x 1/6 = 5/216 +
5/6 x 1/6 x 1/6 = 5/216

The odds of winning £3 are 1/216

1/6 x 1/6 x 1/6 = 1/216

The overall winnings are therefore

(75/216 x $1) + (15/216 x $2) + (1/216 x $3) =

$75/216 + $30/216 + $3/216 = £108/216 = 50 cents.

The house will on average pay out half the stakes and keep half.

No: you forget that if your number doesn't come up at all, you lose your stake. This happens in 125/216 of all cases. Your overall expectation is therefore

(75/216 x $1) + (15/216 x $2) + (1/216 x $3) - (125/216 x $1) = -17/216 x $1

If you play 216 times, you expect to lose 17$.
 
Oh no, I've been completely misunderstood.

Sorry, folks. Let me try again.

If say, you bet on the number 1. And the 3 die show 2, 3 and 4, you lose your bet.

If say, you bet on the numbers 1, 2 and 3 and the 3 die show 4, 5 and 6, you lose all 3 bets.

If say, you bet on the numbers 1, 2 and 3 and the 3 die show 1, 5 and 6, you win the bet on number 1 but lose the other 2.

To all appearances, it looks like a 50-50 thing between the house and the punter. 3 die, 6 numbers on the board - doesn't it look like 50-50?

Forgive me, what I didn't take into account is that this game is completely alien to most of you. Here in Singapore, most would immediately know what I'm talking about.

It does indeed look like 50/50 at first glance. But it isn't. It may help to imagine what will happen if you place six bets at once, one bet on 1, one on 2, one on 3, one on 4, one on 5 and one on 6:

A. If each of the three dice shows a different number, you come out even. Let's say the dice show 1, 2 and 3: you win three dollars (one for each of these numbers) and you lose three dollars for the bets you placed on 4, 5 and 6.

B. If two of the three dice show the same number, you've been had. Let's say the dice show 1, 1 and 2: you win two dollars for the double number 1 and one dollar for the number 2, but you lose four dollars for the four numbers that didn't come up at all. You're down one dollar.

C. If the three dice all show the same number, you're worse off: you win three dollars for that number, but you lose five dollars on the five other bets. You're two dollars down.

The probability of situation B (where you lose 1 dollar) is 15/216 and the probability of situation C (where you lose 2 dollars) is 1/216, in the long run you expect to lose (15/216 + 2 x 1/216) dollars for each dollar you bet. This is the result I already gave in my previous post: you expect a total loss of 17/216 dollars for each dollar you bet.
 
Why would you think it's 50-50?

A good way to see it's not is to imagine betting on all 6 numbers. If all the dice show different numbers you break even - so that's a fair bet. But if two dice are equal, you win 3 and lose 4, so you're down 1, and if three dice are equal you're down 2. Hence there is an advantage for the house.

Say you bet on 1. The odds there will be no 1s showing are (5/6)*(5/6)*(5/6) = 57.9%. That's a pretty big edge to the house. But winning more with more 1s might have offset that, but in this case it won't because the returns aren't as high as they should be to match the odds. Let's add it up:

-5^3/6^3 + 3 5^2/6^3 + 2 3 5/6^3 + 3/6^3 = -0.0787

So a nearly 8% advantage for the house. Betting on more numbers obviously doesn't change anything - all the numbers are equivalent.

The underlying reason this isn't a fair bet is that the odds of getting two 1s are less than half the odds of getting one 1 - so being paid 2 for that isn't an even bet.
 
Last edited:
"Alas", cried Tom, "Now that I know it won't be Dick, it would mean it's either Harry or me who will hang tomorrow. My chances of dying has gone up from 1/3 to 1/2".
Since this is analogous to the Monty Hall problem the only proper phrasing of the correct answer is "Tom should change his name to Harry".
 
It would be the Monty Hall problem if Tom were given the chance to change identies (as far as the executioner is concerned) with one of the other prisoners. If that were the case, should he change? Living is the goat and dying is the car...
Someone always beats you to it on this forum.
 
No: you forget that if your number doesn't come up at all, you lose your stake. This happens in 125/216 of all cases. Your overall expectation is therefore

(75/216 x $1) + (15/216 x $2) + (1/216 x $3) - (125/216 x $1) = -17/216 x $1

If you play 216 times, you expect to lose 17$.
OK I read it that you paid $1 whatever, i.e. you needed 1 number to break even. Hence the average loss of 50 cents a go.
 
I like the dice puzzle! (It might work even better if you go through the incorrect 50-50 solution and ask why it's wrong.)


No: you forget that if your number doesn't come up at all, you lose your stake. This happens in 125/216 of all cases. Your overall expectation is therefore

(75/216 x $1) + (15/216 x $2) + (1/216 x $3) - (125/216 x $1) = -17/216 x $1

If you play 216 times, you expect to lose 17$.
That's correct, but I predict that your explanation will cause confusion. The "- (125/216 x $1)" looks at first sight like a fiddle factor, because if you don't think about it carefully it seems that you've already taken your stake into account.

I think it's easier to see if you start by paying $216 and calculate the expected return, including your stake for each win:
$4*1 + $3*15 + $2*75 = $199


Why would you think it's 50-50?

A good way to see it's not is to imagine betting on all 6 numbers. If all the dice show different numbers you break even - so that's a fair bet. But if two dice are equal, you win 3 and lose 4, so you're down 1, and if three dice are equal you're down 2. Hence there is an advantage for the house.

Say you bet on 1. The odds there will be no 1s showing are (5/6)*(5/6)*(5/6) = 57.9%. That's a pretty big edge to the house. But winning more with more 1s might have offset that, but in this case it won't because the returns aren't as high as they should be to match the odds. Let's add it up:

-5^3/6^3 + 3 5^2/6^3 + 2 3 5/6^3 + 3/6^3 = -0.0787

So a nearly 8% advantage for the house. Betting on more numbers obviously doesn't change anything - all the numbers are equivalent.

The underlying reason this isn't a fair bet is that the odds of getting two 1s are less than half the odds of getting one 1 - so being paid 2 for that isn't an even bet.
My guess is that most punters aren't making the 50-50 mistake (they don't get that far); they probably think something like:
There's a 1 in 6 chance with each die, and I get 3 throws, and then some extra chances of getting 2 or 3 hits, so it's got to be better than even.

Although it's easy to see what's wrong with that, I might not immediately spot the intuitive way to tell whether the house or the punter has the edge. I'd be more likely to work it out by getting the probabilities for 1, 2 and 3 hits and summing the expected payouts (as above), which I might have trouble doing in my head.

Actually I wouldn't bother to do the calculation – I'd assume that a professional gambler knew what he was doing and refuse the bet!
 
Although it's easy to see what's wrong with that, I might not immediately spot the intuitive way to tell whether the house or the punter has the edge. I'd be more likely to work it out by getting the probabilities for 1, 2 and 3 hits and summing the expected payouts (as above), which I might have trouble doing in my head.

You don't need to do that. Since you know betting on 1 is the same as betting on any other number (i.e. all numbers are equivalent) and the bets are independent, the quickest way to evaluate the edge is to consider betting on all 6 numbers. Then you can immediately see the house has an advantage.
 
Last edited:
How is this scenario different from the original post?

Three people are in a jail.
The warden comes in and says “I’m going to eventually kill one, and only one, of you three.”
The warden leaves.
The warden returns and sets one person free, two people remain.
The warden says to them “I’m going to eventually kill one, and only one, of you two.”

In this case the odds of dying clearly change. If given a choice as to when you would like to enter this game, you would want to choose the point with three prisoners left and not two.

Again, if so, how is this different from the original puzzle?
 
Last edited:
Is this confusion over whether it asks if he is right that his odds changed relative to:

the point in time at the beginning of the game.

vs.

the point in time the new information was aquired.
 
Again, if so, how is this different from the original puzzle?

It's different in so far as "Tom" (i.e. whoever talks to the warden in the first scenario) has a one in three chance of being let go half-way through the game.

In the original puzzle, the one asking will not ever be the one who gets named as one of the two survivors.
 
Someone always beats you to it on this forum.
It would be the Monty Hall problem if Tom were given the chance to change identies (as far as the executioner is concerned) with one of the other prisoners. If that were the case, should he change? Living is the goat and dying is the car...
{Ahem...}


OK I read it that you paid $1 whatever, i.e. you needed 1 number to break even. Hence the average loss of 50 cents a go.
Erm ... I don't think they'd get many takers! :D


Why would you think it's 50-50?
Well you might think that if you calculated the payout ignoring the returned stakes. It's true that this part of the payout is exactly half the takings – the trick is to realise that that's not the answer. (The house has an edge because they return less than half the stakes - as you win only 91 times to 125.)

But, as I said, I don't think that's what the punters are doing.


You don't need to do that. Since you know betting on 1 is the same as betting on any other number (i.e. all numbers are equivalent) and the bets are independent, the quickest way to evaluate the edge is to consider betting on all 6 numbers. Then you can immediately see the house has an advantage.
True, it's much simpler; that's what I meant by the intuitive way. But not many people would think to do that – I might not.
 
How is this scenario different from the original post?

Three people are in a jail.
The warden comes in and says “I’m going to eventually kill one, and only one, of you three.”
The warden leaves.
The warden returns and sets one person free, two people remain.
The warden says to them “I’m going to eventually kill one, and only one, of you two.”

In this case the odds of dying clearly change. If given a choice as to when you would like to enter this game, you would want to choose the point with three prisoners left and not two.

Again, if so, how is this different from the original puzzle?
This is how I see it. This is where the confusion is occuring. If he has the ability to change his pick after eliminating 1 of the 3 options, then the odds have changed to 1:2. If he deicdes when there are 3 choices and is told what one of the choices is, but isn't allowed to change his pick, then the odds would still be 1:3. That determining factor is can he change his mind after the pick. Tell me onlytellstruth and I are wrong?
 
This is how I see it. This is where the confusion is occuring. If he has the ability to change his pick after eliminating 1 of the 3 options, then the odds have changed to 1:2. If he deicdes when there are 3 choices and is told what one of the choices is, but isn't allowed to change his pick, then the odds would still be 1:3. That determining factor is can he change his mind after the pick. Tell me onlytellstruth and I are wrong?

In 1/3 of the cases, his chances of being killed change to 0. In the other 2/3 of all cases, he'll have a 1:2 chance of surviving even if not set free in the first round. (1:2 of 2/3, that is.)

In the end, he'll still be killed in 1:3 cases.
 
How is this scenario different from the original post?

Three people are in a jail.
The warden comes in and says “I’m going to eventually kill one, and only one, of you three.”
The warden leaves.
The warden returns and sets one person free, two people remain.
The warden says to them “I’m going to eventually kill one, and only one, of you two.”

In this case the odds of dying clearly change. If given a choice as to when you would like to enter this game, you would want to choose the point with three prisoners left and not two.

Again, if so, how is this different from the original puzzle?

Rasmus has already mentioned an essential difference: in this version, "Tom" may be set free by the warden. The first scenario is asymmetrical: the warden will never name Tom as the person to be set free, even if he is one of the two lucky ones.

In the first scenario, the warden can only choose between naming Dick or Harry. If Dick is to be hanged, he must name Harry. If Harry is to be hanged, he must name Dick. If Tom is to be hanged, the warden will name either Dick or Harry. The proposed solution assumes that the warden is impartial and has an equal chance of naming Dick or Harry in this situation.

In your example, once the warden has set a prisoner free, each of the remaining two can say "that could have been me, but it wasn't". They both have new information about their own situation and each of them can count his chances of survival as being 50% from now on. In the original example, Tom has no new information about himself when the warden tells him that Dick's life will be spared. This information allows him to infer that the probabilities of survival for Dick and Harry have changed (Dick now 0, Harry 2/3), but doesn't change his own chances of survival.
 
This is how I see it. This is where the confusion is occuring. If he has the ability to change his pick after eliminating 1 of the 3 options, then the odds have changed to 1:2. If he deicdes when there are 3 choices and is told what one of the choices is, but isn't allowed to change his pick, then the odds would still be 1:3. That determining factor is can he change his mind after the pick. Tell me onlytellstruth and I are wrong?

It's not whether he can change his pick that is the determining factor, but whether any information is imparted by the release of the prisoner.

In onlytellstruth's example, Tom being released is a possibility. When he isn't, it adds information. In the orignal problem, Tom being told he will be one of the prisoners set free (analogous to being set free in onlytellstruth's example) is not a possibility, so it doesn't add any information.
 
OnlyTellsTruths, ServiceSoon:

It might help to consider the probabilities as masses of 'stuff' (total mass 1), or perhaps piles of tokens that sum to 1, that are re-distributed among the prisoners each time a piece of information is acquired. In either version, each prisoner starts out with a probability of 1/3 (obviously), and the reprieved prisoner's probability collapses to 0 when the reprieve is announced. The difference is in how the newly-available extra probability of 1/3 is distributed between the two remaining prisoners.

In the original version, when one reprieved prisoner is named there is no new information for Tom about his own chances, because he ruled himself out of the warder's answer by being the one to ask the question. (As Michael C says, the warder could not answer "Tom will be reprieved" regardless of whether it was true, and Tom knew this.) So, when Dick's probability collapsed to 0, all of his previous probability attached itself to Harry (I mean, as far as Tom was concerned – remember that probability is a function of what is known, and therefore is generally not the same for people with differing knowledge of the situation).

In the new version, both remaining prisoners have gained (the same) extra information by not being selected when they could have been. The freed prisoner's probability is now shared evenly between the two remaining ones.

As several people have said, the difference between the prisoner versions has nothing to do with the ability (or otherwise) of a prisoner to change his guess. In fact, the prisoner problems have nothing to do with guessing at all – the prisoners (including Tom) correspond to Monty Hall's doors, not to the contestant.
 

ISF - Join now!

Every member here is approved by hand. No bots, no spam, just people who care about evidence and honest debate.

Membership is free!

Create your free account

Back
Top Bottom