Slashdot Mirror
The Tuesday Birthday Problem
An anonymous reader sends in a mathematical puzzle introduced at the recent Gathering 4 Gardner, a convention of mathematicians, magicians, and puzzle enthusiasts held biannually in Atlanta. The Tuesday Birthday Problem is simply stated, but tends to mislead both intuitive and mathematically informed guesses. "I have two children, one of whom is a boy born on a Tuesday. What's the probability that my other child is a boy?" The submitter adds, "Believe it or not, the Tuesday thing is relevant. Well, sort of. It's ambiguous."
13/27
First off, I am a huge Martin Gardner fan and if this puzzle intrigues you and you haven't heard of him, get one of his books.
This problem hinges very greatly on how it is phrased and I think it's more a trick of English converting to statistics than it is a true puzzle. If you were to rephrase this problem as "My first child was a boy born on Tuesday now what are the odds that my next child is a boy?" But they don't. They phrase it as after the fact of both births we are now studying a set of two objects that have been determined prior to me asking the question. This ordering causes the set to be enumerable. Which brings in an interesting piece of information theory to this game. Whenever you say something about one child that is exclusive to that child and that trait is enumerable than you have just affected the outcome of that second child. In the original two childs problem this is just gender. But in the above problem it pairs gender with day of week the child was born on. Now since ordering matters you recognize that whether or not the older or younger child is the one in question creates a different scenario and you can't have twins because only one was born on Tuesday. The article does a good job of explaining this.
The interesting thing is that the answer to this comes down to 13/27. And the larger the enumerable set is of possibilities, the closer this converges to 1/2. If you did this with a specific day of the year, your answer would be 729/1459 which is even closer to 1/2. The general rule is that for a set with N possibilities you would have (N*2 - 1)/(N*4 - 1). Now, what's interesting is if N is unbounded or unenumerable? What if I said "I have two children, one of whom is a boy that likes the number 1835736583. What's the probability that my other child is a boy?" Wouldn't it converge to 1/2?
My work here is dung.
Let's assume that, if I have two children, it is equally probable that they are born as boy+boy, boy+girl, girl+boy, girl+girl. If I have one boy, girl+girl has probability 0, and the other options are equally likely, so they have probability 1/3. If it is known that I have a boy, there is 1/3 probability that the other is also a boy.
X=one boy is born on a tuesday
P(X|boyboy) = 1/7 + (6/7*1/7) = 13/49
P(X|boygirl) = 1/7
P(X|girlboy) = 1/7
P(boyboy) = P(boygirl) = P(girlboy) = 1/3
P(X) = (1/7 + 1/7 + 13/49)/3 = 9/49
Using Bayes's theorem:
P(boyboy|X) = P(X|boyboy)*P(boyboy)/P(X) = 13/49 * 1/3 * 49/9 = 13/27
Which is different from 1/3. So yes, the weekday of birth is significant.
"I have two children, one of whom is a son born on a Tuesday. What is the probability that I have two boys?" .. this is the same as saying "I have just tossed a 10 pence coin and it has come up heads, what is the probability that another coin toss will come up heads?
No, it isn't. It's the same as saying, "I have just tossed a quarter twice, and given that it came up heads at least once, what's the chance it that it came up head both times?"
We can do an analogue, but not in the way you're saying. I.e., "I have just tossed a coin and it came up heads, what's the chance it will come up heads if I toss it again" == "I have just had a kid and it's a boy, if I have another kid, what's the chance it will be a boy" == 50%
If you don't understand the problem, that's fine. It's counterintuitive, these things usually take a second look. If you don't understand the problem and you want to claim the solution is incorrect because you don't get it, well, that's something else entirely.
I don't believe in time. It's a grand conspiracy designed to sell watches.
This is kind of like saying "I flip a coin. What is the chance it lands heads facing up?"
And you say "50%."
And I say, "Incorrect. There is a very small chance it will land balanced perfectly on it's side, so both the chance of heads and the chance of tails is under 50%."
Big apple, new Yorik, undig it, something's unrotting in Edenmark.
Nope.
"I have just tossed a 10 pence coin and it has come up heads, what is the probability that another coin toss will come up heads?"
Is *not* the same as:
"I just tossed two coins and one of them was heads, what is the probability that the other one was heads as well?"
No sig today...
No joke, you break out the SQL.
I did so.
I generated a bunch of SQL records, into a table with two relevant fields, both binary. (Male was indicated by false, female was indicated by true).
Then I counted how many SQL records were returned WHERE gender1 = male OR gender2 = male
Then I counted how many SQL records were returned WHERE gender1 = male AND gender2 = male.
The results? Out of a population of 100,000 records: 24940 male-male, 74893 at least one male. Yielding 1/3 to the third decimal place.
The mistake you made (and it is a mistake, because you claim your experiment would show a result that it empirically does not) is that you failed to count properly. There will always be approximately twice as many records with one boy as there are two boy records.
It's not the BIRTH ORDER that turns this problem into something interesting.
As a side note, in the future, actually run your experiments if it is actually feasible to do so.
WARNING! This girl exceeds the MAXIMUM SAFE standards established by the FDA for BRATTINESS
This is a gambler's fallacy problem. The more tangents you throw at it, the closer you get to .5 (50%), while never reaching it. This is the limit, why? Because there's only two potential outcomes for the other child: boy or girl.
What you (or the website you copied and pasted the ratio from) fail to take into account (and why it's a Gambler's fallacy problem) is that when involving chance, anything that happened in the past is completely irrelevant to future probables. I could roll a die 99 times, and get 6, the probability of getting 100 6's when I've already got 99 6's is still 1 out of 6, not 6^100.
The reason the chi square doesn't come into play here is because it doesn't MATTER the order. Has she said "What is the probability my SECOND-BORN was a boy?" it would be perfectly logical to write the square because the boy who was born on Tuesday could be either the first born or the second born, she never stipulated.
We can say that the boy, who was born on a tuesday, was also a Gemini. Does this change the ratio? No, the probability of having two boys is still 50-50%, because the unknown only has two possible outcomes: boy or girl.
When modding "Informative", please make sure it both has a source and IS actually informative.
No, you're wrong. Look at this. If Mr. Smith has two children, at least one of whom is a boy, it is two times as likely for him to have a boy and a girl than it is for him to have a boy.
Your mistake is in believing that, by virtue of naming one of the boys Peter, the probabilities are magically equalized. They're not. The correct probabilties for your table are:
Peter, Boy = 1/6
Boy, Peter = 1/6
Peter, Girl = 1/3
Girl, Peter = 1/3
Failed to read the article, hey?
You're absolutely correct IF you assume that the man is just describing one of his children. That is the logical interpretation of the problem. However, if someone went out and specifically selected a family with at least one boy, the probability of the other child being a girl is not 50% - the population in question has been artificially depleted of girls because the criteria excluded all two-girl families.
Read the article. Whether the answer is 50% or not depends on the context, and that context is not specified in the problem.
If you meet the person with his son, and he tells you that that son is born on a Tuesday, then the chance of the other kid being a boy is 50%. If you see him as one of the thousands of parents who have two children, at least of which is a boy who's born on a Tuesday, the chance of the other kid being a boy is 13/27.
Close. The situation with coins is this:
I flip a coin twice and record the answer. I repeat this many times. I discard all the pairs where both coins came up tails. I then select a pair where at least one toss came up heads. What is the probability that the other is also heads?
The selection criteria screw up the probabilities. If instead I flip a coin, see that it comes up heads, and ask what the probability of it coming up heads again is, the answer is 50%.
Assuming 50/50 chance of boy or girl and a 1/7 chance of each weekday, let's look at the general population of families with two childs:
B1B1 B1B2 B1B3 B1B4 B1B5 B1B6 B1B7
B2B1 B2B2 B2B3 B2B4 B2B5 B2B6 B2B7
B3B1 B3B2 B3B3 B3B4 B3B5 B3B6 B3B7
B4B1 B4B2 B4B3 B4B4 B4B5 B4B6 B4B7
B5B1 B5B2 B5B3 B5B4 B5B5 B5B6 B5B7
B6B1 B6B2 B6B3 B6B4 B6B5 B6B6 B6B7
B7B1 B7B2 B7B3 B7B4 B7B5 B7B6 B7B7
B1G1 B1G2 B1G3 B1G4 B1G5 B1G6 B1G7
B2G1 B2G2 B2G3 B2G4 B2G5 B2G6 B2G7
B3G1 B3G2 B3G3 B3G4 B3G5 B3G6 B3G7
B4G1 B4G2 B4G3 B4G4 B4G5 B4G6 B4G7
B5G1 B5G2 B5G3 B5G4 B5G5 B5G6 B5G7
B6G1 B6G2 B6G3 B6G4 B6G5 B6G6 B6G7
B7G1 B7G2 B7G3 B7G4 B7G5 B7G6 B7G7
G1B1 G1B2 G1B3 G1B4 G1B5 G1B6 G1B7
G2B1 G2B2 G2B3 G2B4 G2B5 G2B6 G2B7
G3B1 G3B2 G3B3 G3B4 G3B5 G3B6 G3B7
G4B1 G4B2 G4B3 G4B4 G4B5 G4B6 G4B7
G5B1 G5B2 G5B3 G5B4 G5B5 G5B6 G5B7
G6B1 G6B2 G6B3 G6B4 G6B5 G6B6 G6B7
G7B1 G7B2 G7B3 G7B4 G7B5 G7B6 G7B7
G1G1 G1G2 G1G3 G1G4 G1G5 G1G6 G1G7
G2G1 G2G2 G2G3 G2G4 G2G5 G2G6 G2G7
G3G1 G3G2 G3G3 G3G4 G3G5 G3G6 G3G7
G4G1 G4G2 G4G3 G4G4 G4G5 G4G6 G4G7
G5G1 G5G2 G5G3 G5G4 G5G5 G5G6 G5G7
G6G1 G6G2 G6G3 G6G4 G6G5 G6G6 G6G7
G7G1 G7G2 G7G3 G7G4 G7G5 G7G6 G7G7
Each outcome has P = 1/(7*7*4) = 1/196
Let's only look at the families with (at least one) tuesday boy:
B1B2
B2B1 B2B2 B2B3 B2B4 B2B5 B2B6 B2B7
B3B2
B4B2
B5B2
B6B2
B7B2
B2G1 B2G2 B2G3 B2G4 B2G5 B2G6 B2G7
G1B2
G2B2
G3B2
G4B2
G5B2
G6B2
G7B2
Of these 27 families, 13 have another boy. So P = 13/27.
You're right! Pretty amazing, considering the article gives the correct answer and explains it pretty thoroughly. Even when you think: "But wait! Can't you look at it some other way?", the article does just that.
Really, this time it pays to RTFA.
Assuming 50/50 chance of boy or girl and a 1/7 chance of each weekday, let's look at the general population of families with two childs:
B1B1 B1B2 B1B3 B1B4 B1B5 B1B6 B1B7
B2B1 B2B2 B2B3 B2B4 B2B5 B2B6 B2B7
B3B1 B3B2 B3B3 B3B4 B3B5 B3B6 B3B7
B4B1 B4B2 B4B3 B4B4 B4B5 B4B6 B4B7
B5B1 B5B2 B5B3 B5B4 B5B5 B5B6 B5B7
B6B1 B6B2 B6B3 B6B4 B6B5 B6B6 B6B7
B7B1 B7B2 B7B3 B7B4 B7B5 B7B6 B7B7
B1G1 B1G2 B1G3 B1G4 B1G5 B1G6 B1G7
B2G1 B2G2 B2G3 B2G4 B2G5 B2G6 B2G7
B3G1 B3G2 B3G3 B3G4 B3G5 B3G6 B3G7
B4G1 B4G2 B4G3 B4G4 B4G5 B4G6 B4G7
B5G1 B5G2 B5G3 B5G4 B5G5 B5G6 B5G7
B6G1 B6G2 B6G3 B6G4 B6G5 B6G6 B6G7
B7G1 B7G2 B7G3 B7G4 B7G5 B7G6 B7G7
G1B1 G1B2 G1B3 G1B4 G1B5 G1B6 G1B7
G2B1 G2B2 G2B3 G2B4 G2B5 G2B6 G2B7
G3B1 G3B2 G3B3 G3B4 G3B5 G3B6 G3B7
G4B1 G4B2 G4B3 G4B4 G4B5 G4B6 G4B7
G5B1 G5B2 G5B3 G5B4 G5B5 G5B6 G5B7
G6B1 G6B2 G6B3 G6B4 G6B5 G6B6 G6B7
G7B1 G7B2 G7B3 G7B4 G7B5 G7B6 G7B7
G1G1 G1G2 G1G3 G1G4 G1G5 G1G6 G1G7
G2G1 G2G2 G2G3 G2G4 G2G5 G2G6 G2G7
G3G1 G3G2 G3G3 G3G4 G3G5 G3G6 G3G7
G4G1 G4G2 G4G3 G4G4 G4G5 G4G6 G4G7
G5G1 G5G2 G5G3 G5G4 G5G5 G5G6 G5G7
G6G1 G6G2 G6G3 G6G4 G6G5 G6G6 G6G7
G7G1 G7G2 G7G3 G7G4 G7G5 G7G6 G7G7
Why are you assuming that order matters, but boys are interchangable? If we label the child that was mentioned in article as A and the other one as B then B2B2 splits into two cases when A has an older brother and when he has a yonger brother, therefore we get:
B2B1 B2B2 B2B3 B2B4 B2B5 B2B6 B2B7
B1B2
B2B2
B3B2
B4B2
B5B2
B6B2
B7B2
B2G1 B2G2 B2G3 B2G4 B2G5 B2G6 B2G7
G1B2
G2B2
G3B2
G4B2
G5B2
G6B2
G7B2
Of these 28 families 14 have another boy, so P=14/28=1/2
You are neglecting to treat the two children as independent non-exchangeable objects. It is easier to think about if you consider that you have two pets a dog and a cat, each of which can be male or female. Then the enumeration of possibilities:
Male Dog, Male Cat
Male Dog, Female Cat
Female Dog, Male Cat
Female Dog, Female Cat
If you select a family where (at least) one of them is a Male that leaves the following options:
Male Dog, Male Cat
Male Dog, Female Cat
Female Dog, Male Cat
So the chances of both being male are 1/3.
Now moving onto the date question. You have selected a family from families known to have one male pet born on Tuesday (and one dog and cat each). The options are:
Male Cat born Tuesday, Male Dog born Monday
Male Cat born Tuesday, Male Dog born Tuesday *
Male Cat born Tuesday, Male Dog born Wednesday
Male Cat born Tuesday, Male Dog born Thursday
Male Cat born Tuesday, Male Dog born Friday
Male Cat born Tuesday, Male Dog born Saturday
Male Cat born Tuesday, Male Dog born Sunday
Male Cat born Tuesday, Female Dog born Monday
Male Cat born Tuesday, Female Dog born Tuesday
Male Cat born Tuesday, Female Dog born Wednesday
Male Cat born Tuesday, Female Dog born Thursday
Male Cat born Tuesday, Female Dog born Friday
Male Cat born Tuesday, Female Dog born Saturday
Male Cat born Tuesday, Female Dog born Sunday
Male Dog born Tuesday, Male Cat born Monday
Male Dog born Tuesday, Male Cat born Tuesday *
Male Dog born Tuesday, Male Cat born Wednesday
Male Dog born Tuesday, Male Cat born Thursday
Male Dog born Tuesday, Male Cat born Friday
Male Dog born Tuesday, Male Cat born Saturday
Male Dog born Tuesday, Male Cat born Sunday
Male Dog born Tuesday, Female Cat born Monday
Male Dog born Tuesday, Female Cat born Tuesday
Male Dog born Tuesday, Female Cat born Wednesday
Male Dog born Tuesday, Female Cat born Thursday
Male Dog born Tuesday, Female Cat born Friday
Male Dog born Tuesday, Female Cat born Saturday
Male Dog born Tuesday, Female Cat born Sunday
* Note that I enumerated the case where both are males born on Tuesday twice. These are redundant and one must be discarded else I will double count that situation. After doing so there are 13/27 cases where both are males.
Notice that if you ignored the fact that one was a dog and the other was a cat you would have merged the two lists, ending up with your original list, and double counting the case where both are boys born on Tuesday.
In other words your mistake is that you assumed you had been given the sex and birth date of child A, and enumerated the sex and birth date of child B. However, you don't know the sex and birth date of child A or B, just that one of child A or B have that sex and birthdate. That is a subtly different problem.
So the order that they are born in is irrelevant, but keeping track of the fact that they are the two unique items while enumerating the cases is vital (and older and younger is a simple label to use while doing so).