A family has two children. At least one of them is a boy. What is the probability that both are boys?
Most people answer 1/2 instinctively. The “right” answer, under standard assumptions, is 1/3. The interview signal is whether the candidate sees why their first instinct is wrong, and whether they can articulate what assumptions push the answer one way or the other. Quant interviewers and statistics-heavy tech rounds love this question because it cleanly separates candidates who think probabilistically from candidates who think in heuristics.
The setup
Assume that births are independent and that boys and girls are equally likely. Before any information, the four equally-likely outcomes for two children, ordered by birth, are:
- BB (older boy, younger boy) — probability 1/4
- BG (older boy, younger girl) — probability 1/4
- GB (older girl, younger boy) — probability 1/4
- GG (older girl, younger girl) — probability 1/4
“At least one is a boy” eliminates GG. Of the three remaining cases, only one (BB) has both being boys. So:
P(both boys | at least one boy) = 1/3.
Why most people say 1/2
The intuitive (wrong) answer treats the problem as: “you know one child is a boy, so the other child is independently a boy or a girl with equal probability”. This argument would be correct if the question were phrased differently — “the older child is a boy, what is the probability the younger is a boy?” In that framing, the answer is 1/2, because conditioning on a specific child eliminates GG and BG (since the older being boy rules out older=girl), leaving BB and GB, half of which is BB.
The 1/3 versus 1/2 distinction comes down to whether you condition on “a specific child is a boy” or “at least one of the children is a boy”. Those are different events with different posterior probabilities. The classic version of the puzzle uses “at least one”, which is why the answer is 1/3.
The Tuesday boy
The Tuesday boy variation, popularized by Gary Foshee at the 2010 Gathering for Gardner conference, is what made the puzzle famous beyond textbooks. The setup adds a date:
“I have two children. One is a boy born on a Tuesday. What is the probability I have two boys?”
The answer is approximately 13/27 ≈ 0.481. Almost exactly 1/2 — but not 1/2, and the difference matters.
To see why, enumerate the day-of-week-and-sex outcomes for two children. There are 14 possibilities per child (7 days × 2 sexes), so 14 × 14 = 196 ordered pairs of children. The condition “at least one boy born on Tuesday” cuts this down. Each child can independently be a “Tuesday boy” or not.
Number of pairs where the older child is a Tuesday boy: 14 (any second child). Number of pairs where the younger child is a Tuesday boy: 14 (any first child). Subtract the double-counted “both Tuesday boys”: 1.
Total pairs with at least one Tuesday boy: 14 + 14 − 1 = 27.
Of these 27, how many have both children being boys? An older Tuesday boy with any boy younger child: 7. A younger Tuesday boy with any boy older child: 7. Subtract the double-count of both being Tuesday boys: 1. Total = 13.
P(both boys | at least one Tuesday boy) = 13/27 ≈ 0.481.
Adding “Tuesday” — a piece of information apparently unrelated to the sex of the children — moves the answer from 1/3 (no day specified) toward 1/2 (specific day given). The reason is that “Tuesday” makes the boy more identifiable, partially turning “at least one is a boy” into “this specific boy is a boy”, which is the conditioning that gives 1/2 rather than 1/3.
Why this question is famous
The Tuesday boy is the cleanest demonstration in the interview canon of how Bayesian conditioning depends on exactly what information you observe and how it was selected. The answer is not 1/3 (the no-extra-info case) and not 1/2 (the specific-child case) but a precise fractional value in between, and the value depends on the cardinality of the discriminating attribute. If you replaced “Tuesday” with “born in 2014” (a more specific identifier), the answer would be even closer to 1/2. If you replaced it with “born in some year” (no extra info), the answer reverts to 1/3.
This is the puzzle that makes statisticians and quants careful about phrasing in real-world problems. A question like “what is the probability of disease X given symptom Y?” depends critically on how the symptom was observed — was it volunteered by the patient (selection bias) or detected by a screening test (no selection bias)? Different observation processes give different posteriors for the same observed event. The Tuesday boy is the toy version of that real concern.
Variations interviewers ask
- “At least one is a boy born on a sunny day.” Probability of two boys is roughly 1/3, because “sunny day” is roughly half of all days and there are far fewer “sunny day” boys than total boys, so the conditioning weakens. The exact answer depends on the day proportion.
- “My older child is a boy.” Answer is 1/2, because conditioning on a specific child kills the BG/GB asymmetry.
- “I have two children. I tell you one is a boy by introducing him to you.” The answer depends on the introduction protocol. If the parent always picks a boy when there is one, the answer is 1/3. If the parent picks a child uniformly at random and reports its sex, the answer is 1/2 (the “Monty Hall conditioning” subtlety).
- The full Tuesday Boy with weighted days. If “born on Tuesday” is replaced with a non-uniform attribute (rarer days have a different prior), the conditional probability moves further from 1/3.
The selection-protocol subtlety — “what is the rule by which the information was given to you” — is the deepest version of the question. A polished candidate raises it unprompted, even before computing anything, because without that detail the problem is genuinely ambiguous.
What it tests
Quant interviewers ask the two-children paradox and the Tuesday boy because the underlying skills are job-essential:
- Conditioning awareness. Recognizing that “at least one boy” and “this specific child is a boy” are different events. Many quant problems involve choosing the right conditioning, and getting it wrong gives a fractionally wrong answer that compounds.
- Selection-protocol thinking. Understanding how information is generated, not just what information is given. This is core to thinking about market data, sampling, and causal inference.
- Discrete enumeration under pressure. Computing 13/27 in a conversation requires keeping a clean accounting of cases without losing the inclusion-exclusion correction. A candidate who does it in their head signals strong combinatorial fluency.
Is it asked in 2026?
Yes, regularly at quant interviews. Jane Street, Citadel, Two Sigma, Optiver, and SIG all use variants of the Tuesday boy puzzle, often with twists designed to expose candidates who have memorized the answer to the canonical version. The standard follow-up — “what changes if I told you about him by picking a child uniformly at random rather than always picking a boy?” — separates candidates who have actually thought about the problem from those who have just looked up the answer.
In tech interviews proper, the question is rare. It still appears occasionally in ML/data science loops where the interviewer wants to test Bayesian intuition, and in some statistics-flavored data engineering interviews. But the dominant tech tradition is around coding and system design, not probability puzzles.
Frequently Asked Questions
Why is the answer 1/3 and not 1/2?
“At least one is a boy” rules out only the girl-girl case, leaving boy-boy, boy-girl, and girl-boy as the three equally likely cases. Only boy-boy has both being boys, giving 1/3. The 1/2 intuition implicitly conditions on a specific child, which is a different event.
Where does 13/27 come from in the Tuesday boy?
Of the 196 sex-and-day combinations for two children, 27 contain at least one Tuesday boy. Of those 27, exactly 13 have both children being boys (count 7 with older Tuesday boy, 7 with younger Tuesday boy, minus 1 for the double-counted case where both are Tuesday boys).
Does the answer depend on selection bias?
Yes, deeply. The 1/3 answer assumes the information was given to you in a way that does not depend on a specific child. If the parent picks a child uniformly at random and tells you its sex, the answer becomes 1/2. The phrasing of the puzzle hides this assumption, which is why it is famous.
Is this a “trick question”?
No, it is a real Bayesian conditioning question. The result is correct under standard assumptions; the surprise is just that intuition fails. Calling it a trick is what people who got the wrong answer say.
What is the cleanest way to explain it in an interview?
Enumerate the four equally-likely birth outcomes BB, BG, GB, GG. State that “at least one boy” eliminates GG. Of the remaining three, only one is BB. So 1/3. Then note that this answer depends on the conditioning protocol and offer the 1/2 variant as a sanity check.