Boy or Girl Paradox Calculator
A family has two children, no twins, about whom nothing else is known. You learn that at least one is a boy. What is the probability both are boys? The answer that feels right — one half — is wrong, and the answer that is right — one third — has started arguments among mathematicians for seventy years. This calculator enumerates the possibilities for each classic version of the puzzle in full, so the counting can be inspected openly rather than debated: choose what you learned, and see exactly which families survive the information, and in what proportion.
The Paradox, Stated Honestly
Gardner popularised the puzzle in 1959 and immediately had to clarify it, because the paradox is not in the arithmetic — it is in the question. “At least one is a boy” can arrive as information in two very different ways. If a family was selected because it satisfies the condition (you surveyed two-child families and kept those with a boy), then the surviving families are BB, BG and GB in equal numbers, and BB is one third. If instead you met one particular child — the one who answered the door — and he is a boy, then his sibling is a fresh coin flip and two boys is one half. Both answers are correct answers to different questions wearing the same words, which is why the argument has survived seventy years of intelligent people talking past one another in perfect good faith. This calculator’s five scenarios are the standard map of the territory, and its counts show the machinery rather than asking you to trust it.
Counting the Worlds
Order the children (older, younger) and the families enumerate cleanly: BB, BG, GB, GG — four equally likely, since each birth is an independent near-even split. Now apply each information state as a filter and count the survivors, which is all conditional probability ever is. At least one boy removes GG: three families remain, one of them BB → 1/3. Older child is a boy removes GB and GG: two remain, one BB → 1/2. At least one girl mirrors the first filter: GG is 1/3. The calculator’s default scenario is the famous one, and its lesson generalises beyond nurseries: conditioning on a property of the set (at least one…) weights the worlds differently from conditioning on a property of an identified member (this one…), and confusing the two is the single most fertile source of wrong probabilities in applied work — from medical screening to the Monty Hall problem, this hub’s other famous resident.
The Tuesday Child: Irrelevant Details That Aren’t
The scenario connoisseurs savour: you learn at least one child is a boy born on a Tuesday. Surely the weekday is irrelevant decoration? It is not — it moves the answer to 13/27 ≈ 48.1%, almost all the way to one half. The counting: each child has 14 equally likely states (sex × weekday); 196 family combinations; those containing at least one Tuesday-boy number 27 (14 with the Tuesday-boy as one specified child, plus 13 more with him as the other, the double-counted TT family counted once); and in 13 of those 27 the other child is also a boy. The irrelevant detail works because it nearly identifies a child — a Tuesday-born boy is a rare, specific creature, so hearing about him is almost like meeting him, and meeting him (scenario five) gives exactly 1/2. Information theory in a nursery rhyme: the more specific the clue, the closer the probability slides toward the identified-child answer.
Assumptions Under the Counting
The enumerations presume the textbook demography: each child independently a boy or girl with probability 1/2, no twins, no selection effects in which families you hear about. Reality files its usual amendments — the human sex ratio at birth runs near 51.2% boys, which nudges the first scenario’s answer only slightly: P(BB | at least one B) = p² / (p² + 2p(1−p)) with p = 0.512 gives ≈ 0.344 rather than 0.333 — and the “how did you learn it” caveat swamps such refinements anyway. The calculator keeps the clean 1/2 births because the paradox is about the structure of information, not obstetrics; the sensitivity check above shows the structure survives realistic numbers with its surprise intact.
Mr Smith and Mr Jones: The Original Pairing
Martin Gardner’s 1959 presentation paired the scenarios deliberately, and the pairing remains the cleanest teaching form. Mr Smith has two children, at least one a boy: both boys at 1/3. Mr Jones has two children, the elder a boy: both boys at 1/2. Gardner himself, in a follow-up, conceded the first problem’s wording was ambiguous — the answer depends on how the “at least one” was ascertained — and that concession, from the puzzle’s own author, is the mature moral: probability questions are questions about an information-gathering protocol, and a problem that doesn’t specify its protocol has not finished specifying itself. The calculator’s scenario menu is, in effect, the five protocols the seventy years of argument have taught us to distinguish.
The Coin Translation, for Skeptics
Strip the nursery away and the paradox survives translation into pure chance, where nobody’s intuition about siblings interferes. Flip two coins and hide them. Told “at least one is heads”, the probability of two heads is 1/3 — the worlds HH, HT, TH survive the filter and HH is one of three. Told “the left coin is heads”, it is 1/2. Anyone can verify the first claim empirically in ten minutes of filtering flips (discard the double-tails, tally the rest), and the exercise has converted more skeptics than any proof — the calculator’s enumeration, performed by hand, at kitchen-table scale. Children were only ever the memorable costume; the filter is the paradox, and coins wear it just as well.
Related Tools
More from the Statistics Calculators hub: Bertrand’s Box Paradox — the same trap with coins and boxes — and the Joint Probability Calculator for the conditional machinery underneath.
Frequently Asked Questions
Why is the answer 1/3 and not 1/2?
Because “at least one boy” does not pick out a child — it filters families. The four ordered family types BB, BG, GB, GG are equally likely; the filter removes only GG; BB is one of the three survivors. The 1/2 intuition smuggles in a specific boy (“this one is a boy, what about the other?”), which is a different, stronger information state — scenario two and five territory — where 1/2 is genuinely correct. The calculator’s counts let you watch the filter work rather than take sides in the argument.
Does the birth order really matter?
It matters to the counting because it makes BG and GB distinct worlds, and both survive the “at least one boy” filter — the mixed families carry double weight against BB’s single world. Ignore order and the family types look like three (two boys, mixed, two girls), which tempts the wrong denominator; restore the order and the 1/3 is unavoidable. Much of elementary probability is exactly this: enumerating the worlds finely enough that they become equally likely, then counting honestly.
What if I learn the information by meeting a child?
Then you are in scenario five, and the answer is 1/2 — the met child is certainly a boy, and his sibling’s sex is an unconditioned coin flip. This is why the paradox’s resolution is always about the protocol: information gathered by filtering a population (“of families with at least one boy…”) differs from information gathered by sampling a member (“this child I met…”), even when the resulting sentence is identical. Surveys, studies and screening programmes are protocols; the sentence is never the whole evidence.
Is the Tuesday detail a trick?
It is the opposite of a trick — it is a measurement of how identifying a clue is. A rarer specified trait (born on a Tuesday, born on June 3rd, left-handed) pins the information to one child more completely, sliding the answer from 1/3 toward 1/2; the limit of a perfectly identifying clue is exactly the met-child case. The calculator includes it because it converts a bar-room paradox into a genuine instrument: probability as a function of how information was specified.
What does this have to do with real statistics?
Everything conditional wears this shape. The false-positive paradox on this hub is the same error in clinical dress (conditioning on a set property versus a member’s). Bertrand’s box and Monty Hall are the same filter confusion with props. Even survey non-response and selection bias are protocol effects: the population that survives your filter is not the population you meant to describe. The nursery is simply where the trap is cheapest to see.
What if there are three children?
The method outlives the example: enumerate 2³ = 8 ordered families, filter, count. “At least one boy” among three children leaves seven families, of which exactly one is BBB — a 1/7 chance all three are boys, against the 1/4 the identified-child intuition might offer (eldest a boy → remaining two both boys at 1/4). The fan of cases is mechanical once the worlds are listed, which is the calculator’s closing sermon: when a probability feels paradoxical, count worlds, in order, without mercy.