Re: A simple betting problem

From: Hal Finney (hal@finney.org)
Date: Wed Apr 30 2003 - 13:59:10 MDT

  • Next message: Terry W. Colvin: "FWD (PvT) Re: The bottom two-thirds of a cosmological iceberg ?"

    Eliezer writes:
    > Here's another trick question that only the true Bayesian will
    > resolve. You meet a mathematician. "How many children do you have?"
    > you ask. "Two," he replies, "and at least one of them is a boy."
    > What is the probability that they are both boys?
    >
    > I would *not* answer 1/3.

    Rafal replies:
    > 1/2 ?

    The answer depends very much on your prior beliefs, as it should in a
    good Bayesian problem. What do you think the mathematician's motives
    are in asking this question? What might he have asked differently in
    other circumstancs? How equal-minded is he with respect to the sexes?

    Depending on the answers to these questions, I could see the answer
    plausibly ranging between 1/2 and 1/3.

    The "classic" answer to this puzzle is 1/3. The reasoning is that if
    all we knew was that he had two kids, they could equally be girl-girl,
    girl-boy, boy-girl, or boy-boy. Once he tells you that at least one
    is a boy, that elminates girl-girl, leaving the remaining three with
    equal probability. Therefore the chance that he has boy-boy is 1/3.

    The problem is that this reasoning assumes that he would pose the puzzle
    in exactly these terms for any of the last three cases. In particular,
    if he has a boy and a girl, you have to assume that he would definitely
    have said, "...and at least one of them is a boy." In other words,
    you have to assume that he is a male chauvinist who views boys as more
    important than girls.

    If on the other hand he is a sexual egalitarian, then if he had a boy and
    a girl he might have said, "...and at least one of them is a girl." If we
    assume that in those cases there was a 50-50 chance that he might have
    said "boy" or "girl", that gives us stronger reason to believe in boy-boy.
    Only with boy-boy would he have certainly have said "boy"; with boy-girl
    or girl-boy there was only a 50% chance he would have used that word. This
    reasoning leads to an overall chance for boy-boy of 1/2.

    Or you could make other assumptions; maybe he is more likely
    to pose the puzzle if he does have two boys, since he knows that
    mathematically-trained acquaintances will be likely to guess wrong.
    In that case the probability might be even greater than 1/2. It all
    depends on your priors.

    Hal



    This archive was generated by hypermail 2.1.5 : Wed Apr 30 2003 - 14:10:35 MDT