The boy or girl paradox calculator above computes the probability that a family of two children has two girls, under five different accounts of how you came to know that at least one of them is a girl. It exists because the famous puzzle has no single answer, and pretending otherwise is the reason it has generated more argument than almost any other elementary probability question.
Arb Digital publishes it as a companion to our other conditional-probability tools. The point of the page is not to hand you a number. It is to make visible the step that almost every retelling of the puzzle skips: the moment where a fact about the world gets turned into evidence, which requires you to say what would have happened in the cases you did not observe. Our conditional probability calculator handles the general machinery; this page applies it to the one problem where the machinery's hidden assumption does all the work.
What This Boy or Girl Paradox Calculator Does
It enumerates the four equally likely birth-order combinations — girl-girl, girl-boy, boy-girl, boy-boy — and then applies a conditioning rule that depends on which sampling story you select. The five stories are the ones that appear in the literature: birth order known, family screened for at least one girl, a single child observed at random, the weekday variant, and a general mentioning bias that contains the others as special cases.
It also lets you vary the probability that any one child is a girl, so the arithmetic does not silently assume an exact coin flip, and it lets you set the rarity of the extra detail in the weekday version. The bars plot all five readings together, which is the fastest way to see how far apart they are and under what conditions they converge.
The fraction shown in the fourth grid item is the exact rational answer where one exists at the standard settings — one half, one third, thirteen twenty-sevenths — rather than a rounded decimal. Change the probability of a girl away from 50 percent and it reports a decimal instead, because the tidy fractions are artefacts of the symmetric case.
How to Use It
- Pick the sampling story that matches how the information actually reached you. This is the only choice that changes the structure of the answer.
- Leave the girl probability at 50 for the textbook version, or adjust it to see how much the classic fractions depend on perfect symmetry.
- Set the weekday rarity if you are exploring the Tuesday variant. Seven is the classic value.
- Set the mentioning bias for the last reading and watch the answer sweep continuously from one third to one half.
- Compare the bars rather than fixating on the headline. The spread between them is the actual content of the puzzle.
The Formula and How It Is Calculated
Write p for the probability that a child is a girl. The four ordered outcomes have probabilities p² for two girls, p(1−p) each for the two mixed orders, and (1−p)² for two boys. Every reading is then an application of the conditional probability definition, P(A given B) = P(A and B) divided by P(B), which Wolfram MathWorld sets out on its conditional probability page.
For the elder-child reading, the condition fixes one specific child, so the other is untouched and the answer is simply p — one half in the symmetric case. For the screened reading, the condition removes only the boy-boy outcome, leaving three equally likely possibilities of which one is girl-girl: the answer is p² divided by (1 − (1−p)²), which is p/(2−p), or exactly one third when p is a half.
The mentioning-bias reading generalises both. Let r be the probability that a parent of one boy and one girl mentions the girl. Then the answer is p² divided by (p² + 2p(1−p)r). Setting r to 1 gives the one-third answer, because a mixed family always mentions the girl; setting r to a half gives exactly one half, because a mixed family is only half as likely to produce the sentence you heard. Every intermediate value is a legitimate model of a real informant, and there is nothing in the puzzle's wording that pins r down.
Why One Third and One Half Are Both Defensible
The one-third answer is correct when the family was selected because it contains at least one girl — for instance, if you drew a family at random from a register of all two-child families with at least one daughter. In that world, "at least one girl" is a screening criterion applied to the population, boy-boy families never appear, and the three surviving combinations are equally likely.
The one-half answer is correct when a child was selected and observed, and the sentence describes that child. If you meet one of the two children in the street and she is a girl, you have learned about one specific child, and the other one is still an untouched coin flip. The sentence "at least one is a girl" is a true description of what you now know, but it is not the evidence you actually received, and using it as the evidence throws away the fact that a particular child was sampled.
Both readings are consistent with the English sentence, which is precisely the difficulty. Martin Gardner, who popularised the problem in Scientific American in 1959, later acknowledged that the second phrasing was ambiguous and that both answers could be defended depending on how the family had been chosen. Peter Lynch's paper "The Two-Child Paradox: Dichotomy and Ambiguity", published in the Irish Mathematical Society Bulletin, works through the competing formulations and argues that the ambiguity is genuine rather than a failure of reasoning by one side.
The Tuesday Variant and Why It Is So Strange
Change the sentence to "at least one is a girl born on a Tuesday" and the screened answer moves from one third to thirteen twenty-sevenths — about 48.1 percent, startlingly close to a half. A detail that appears to be irrelevant to the sex of the second child has moved the answer by fifteen percentage points.
The mechanism is a counting effect, not a causal one. The extra detail makes it harder for a family to satisfy the condition, and it penalises two-girl families less than mixed families in relative terms, because a two-girl family has two chances to produce a Tuesday girl while a mixed family has only one. As the detail gets rarer — raise the weekday number in the tool to 100 or 1,000 — the screened answer creeps ever closer to one half, because the condition is increasingly likely to be satisfied by exactly one identifiable child, which is the random-child reading in disguise.
The crucial observation is that the effect disappears entirely under the random-child reading. If you met a girl and she happens to mention she was born on a Tuesday, the answer stays at one half whatever the weekday. Select that reading in the tool and change the weekday number: the headline does not move. The Tuesday twist is a property of screening, not of the information itself, and that is the sharpest available demonstration that the paradox is about the sampling procedure.
Real Births Are Not Symmetric Either
The textbook problem assumes that each child is a girl with probability exactly one half and that the two children are independent. Neither holds exactly. The human sex ratio at birth is slightly boy-leaning, and the sexes of siblings within a family are not perfectly independent for reasons that are still debated. Set the girl probability to 48.7 percent in the tool and the screened answer moves from 0.3333 to about 0.3219 — a small shift, but enough that the tidy fraction is a modelling choice rather than a fact.
Twins break the model more seriously, because identical twins are always the same sex and the two children are then not independent at all. So does anything that made the family visible to you in the first place — a girls' school register, a mother-and-daughter event, a survey of families with daughters. Each of those is a screening rule with its own probability of admitting a mixed family, which is exactly what the mentioning-bias parameter represents. Our binomial distribution calculator and coin flip probability calculator both assume the independence this problem quietly relies on.
The Family of Puzzles This Belongs To
Three classic puzzles turn on the same hinge, and it is worth seeing them together. In Bertrand's box paradox, the answer depends on the fact that a box is chosen and then a ball is drawn from it, rather than a ball being chosen directly; our Bertrand's box paradox calculator works through that enumeration. In Bertrand's paradox about random chords, three different methods of drawing a chord "at random" give three different answers, and our Bertrand's paradox calculator computes all three; there is no agreed correct one, because "at random" is not a specification.
This page is the third. In all three, an English phrase that sounds like a complete description of a random process turns out to describe several different processes, and the arithmetic is only well defined once you pick one. That is a genuinely useful lesson well beyond puzzles: any statistic derived from data that were collected under an unstated selection rule has the same problem, and the rule usually matters more than the sample size.
For the discrete updating logic behind all of them, our Bayes theorem calculator lets you set the likelihoods explicitly, which is a good way to see that the likelihood — the probability of hearing what you heard under each hypothesis — is the term the puzzle leaves unspecified. Our birthday paradox calculator covers a puzzle that surprises for a different reason: there the answer is uncontested and only the intuition is wrong.
Arb Digital asks how the data were selected before reading anything into them, because the selection rule usually matters more than the sample size.
Browse All Free Tools Talk To Our TeamCommon Mistakes to Avoid
- Insisting one answer is simply correct — the sentence is ambiguous, and one half and one third are each right under a clearly statable sampling procedure.
- Forgetting that girl-boy and boy-girl are separate outcomes — treating "one of each" as a single case makes the three combinations look equally likely when they are not.
- Conditioning on the wrong event — if a specific child was observed, the evidence is about that child, not the weaker statement that at least one is a girl.
- Assuming the Tuesday detail is irrelevant — under screening it changes the answer, because it changes how many families satisfy the condition and by how much.
- Taking the tidy fractions as facts — one third and thirteen twenty-sevenths depend on an exact 50 percent sex ratio and on independence between siblings.
Related Free Tools From Arb Digital
Work through the general case with the conditional probability calculator, update a belief explicitly with the Bayes theorem calculator, build event probabilities from scratch with the probability calculator, compare a puzzle whose answer is uncontested with the birthday paradox calculator, or check the odds of a repeated trial with the dice probability calculator. The full free online tools hub lists every probability tool we publish.
Frequently Asked Questions
Both, under different sampling procedures. One third applies when the family was screened for having at least one girl. One half applies when a specific child was observed and turned out to be a girl.
Because girl-boy and boy-girl are two separate equally likely outcomes. Knowing that the elder child is a girl eliminates two of the four, while knowing only that one of them is a girl eliminates just one.
Under screening it makes the condition harder to satisfy, and it penalises mixed families more than two-girl families because a two-girl family has two chances to produce a Tuesday girl. Under random-child observation it changes nothing.
The probability that a parent with one child of each sex chooses to mention the girl. Setting it to one hundred percent gives the one-third answer and setting it to fifty percent gives one half, so it contains both classic readings.
Yes. After posing it in Scientific American in 1959 he acknowledged that the second phrasing could be read in more than one way and that both answers were acceptable depending on how the family had been chosen.
Slightly. The tidy fractions assume each child is a girl with probability exactly one half. Entering a realistic figure a little below fifty percent moves the screened answer down by about one percentage point.
The birthday problem has one uncontested answer that merely feels wrong. This problem has several correct answers, and which one applies depends on information the question does not supply.
This page explains a probability puzzle for educational purposes. It models sexes as independent events with a fixed probability, which real birth data only approximates.
