How many people are needed before a shared birthday becomes more likely than not? Many people guess well over a hundred. Under the simple model of 365 equally likely birthdays and independent birth dates, the answer is just 23 people. The probability of at least one shared month-and-day birthday is about 50.73%. This is the birthday paradox: a surprisingly high collision probability arising from the number of pairs in a group.
This Mathematics in Bukit Timah guide explains birthday matching, complementary probability, permutations, the multiplication rule, expected matching pairs, approximation, independence, uneven birthday distributions, hash collisions and the difference between likely and guaranteed. Every group in this article is imaginary. It is not a record of Bukit Timah pupils, tutor classes, school enrolments or real birthday data. We make the simplified probability assumptions explicit instead of claiming actual birthdays are uniformly distributed.
Quick answer: For n independently sampled birthdays, each uniformly drawn from 365 possible month-and-day labels, P(at least one shared birthday)=1−∏ from k=0 to n−1 [(365−k)/365] when 0≤n≤365. At n=23 this is approximately 0.5073. With 366 people a match is certain by the pigeonhole principle, but the 50% probability threshold is crossed much earlier.
Article contents
- Why twenty-three feels surprising
- Bukit Timah is not the source of the birthday data
- State the simplified sample space
- What is a collision?
- Worked example 1: a group of two
- Why calculating no match is easier
- The first person’s birthday is unrestricted
- The second person must avoid one date
- The third person must avoid two dates
- Worked example 2: a three-person match
- Build the product for n people
- The full collision probability
- Worked example 3: six people
- Twenty-three participants give 253 pairs
- Worked example 4: why pair count grows quickly
- The threshold calculation
- Important group sizes
- Why one match is likely but not guaranteed at twenty-three
- The birthday paradox has a separate certainty threshold
- Worked example 5: why 365 people do not guarantee a match
- One named person’s birthday is another problem
- Worked example 6: a named person’s chance at twenty-three
- How event definitions change the denominator
- Expected number of matching pairs
- Worked example 7: expected pairs at twenty-three
- Why the expected count can exceed the collision probability
- Worked example 8: one triple produces three pairs
- A useful approximation for the no-match probability
- Worked example 9: approximate the twenty-three-person result
- The approximate fifty-percent threshold
- The square-root rule
- Worked example 10: hypothetical thousand categories
- Why uniformity matters
- General unequal birthday probabilities
- Worked example 11: unequal two-category labels
- Independence is a separate assumption
- Worked example 12: a deliberately dependent case
- Simulating collisions is not the same as proving the exact formula
- Worked example 13: ten simulated groups
- A probability is not a personal forecast
- Collisions appear in computer identifiers too
- Worked example 14: four equally likely labels
- A complete enumeration for three labels
- Worked example 15: use the small enumeration
- Counting without replacement changes everything
- Worked example 16: replacement versus no replacement
- The complement should be explicitly stated
- Why adding pair probabilities is not exact
- Worked example 17: a too-large naive sum
- Different collision questions demand different formulas
- Worked example 18: exactly one match among three
- A triple also creates pair collisions
- Worked example 19: verify the three-person decomposition
- The actual birthday distribution may be seasonal
- A leap-day model needs explicit probabilities
- Worked example 20: uniform 366-day thought experiment
- Birth-year equality is a different condition
- A collision probability increases with group size
- Worked example 21: first two group sizes
- The probability is not linear in group size
- Worked example 22: compare pair counts
- A probability of ninety percent is still not certainty
- The distinction from pigeonhole guarantees
- A useful arithmetic implementation
- Worked example 23: calculate through three steps
- Floating-point calculations still need care
- Common error 1: confuse one specified birthday with any pair
- Common error 2: add pair probabilities directly
- Common error 3: claim twenty-three people guarantee a match
- Common error 4: forget independence
- Common error 5: confuse 365 and 366 labels
- Common error 6: count pairs as n squared
- Common error 7: treat an expected pair count as the event probability
- Common error 8: infer real classroom dates from model probabilities
- Primary readiness: months and simple matching
- Lower Secondary: multiplication and complementary probability
- Advanced probability and syllabus boundaries
- A twenty-minute family diagnostic
- Seven days of independent practice
- Practice questions with checked answers
- Frequently asked questions
- Official context and related Mathematics in Bukit Timah
Why twenty-three feels surprising
The wrong intuition often compares a group with one specific person’s birthday.
Yet the actual question permits a match between any two members. A room of twenty-three people contains many pairs to compare.
The number of possible unordered pairs is 23×22/2=253. Those 253 potential relationships make a collision plausible even though any particular pair has only a 1/365 matching probability under the model.
Bukit Timah is not the source of the birthday data
Imagine a fictional gathering somewhere near Beauty World or Sixth Avenue with twenty-three anonymous participants.
We do not know their dates of birth or whether any two really match. Nothing in the article predicts an actual birthday coincidence for a named class.
The model provides a probability before observing birthday labels. Once a group’s birthdays are known, whether a match exists becomes a factual count rather than a probability prediction about unknown dates.
State the simplified sample space
We ignore February 29 and use 365 possible month-and-day combinations.
Each person is assigned one date label, with all 365 labels equally likely. Assignments are independent across people.
These are convenient teaching assumptions. Real birth dates may vary by season, geography and other factors, and people in a chosen group need not be independent samples.
What is a collision?
A collision occurs when at least two different people have the same birthday label.
It does not require three matches, identical birth years or exactly two people on one date.
If five people happen to share a date, the event “at least one collision” has still occurred. The event is a simple yes-or-no property of the entire group.
Worked example 1: a group of two
The first person may have any birthday.
For the second person to match, their label must equal the first person’s specific date.
Under the uniform independent model, P(match)=1/365≈0.274%. Thus two people are very unlikely to match, even though large-group collision probability will increase rapidly.
Why calculating no match is easier
The event “at least one match” can occur in many overlapping ways.
For example, people A and B might match, or A and C might match, or both events might happen. Adding those probabilities naively double-counts overlapping outcomes.
Instead calculate the complement: every person has a different date. The multiplication rule makes that probability easy.
The first person’s birthday is unrestricted
For the first fictional person, any of the 365 date labels is acceptable.
The probability that this person’s date creates no collision is one because no earlier person exists.
This is the first factor in the no-collision product. Starting the calculation with 364/365 for the first person would omit one legitimate degree of freedom.
The second person must avoid one date
After one date has been used, the second person must choose among the other 364.
Conditional on the first birthday, the probability of avoiding a collision is 364/365.
The first two people therefore have distinct birthdays with probability 1×364/365.
The third person must avoid two dates
If the first two people have different birthdays, two date labels are already occupied.
The third must choose one of the remaining 363 labels.
Conditional probability of success is 363/365. Thus three distinct dates have probability (364/365)(363/365).
Worked example 2: a three-person match
For three people, no-match probability is (364×363)/365².
The collision probability is 1−(364×363)/365².
Numerically it is about 0.008204, or 0.8204%. The answer is larger than 1/365 because three different pairs could match.
Build the product for n people
For n≤365, the successive conditional no-match factors are
365/365, 364/365, 363/365, …, (365−n+1)/365.
Multiplying gives P(no matches)=365!/[365ⁿ(365−n)!].
The factorial expression and the successive-factors expression are equivalent. In practice the product is easier to understand and often easier to compute safely.
The full collision probability
Use the complement rule:
P(match)=1−P(no matches).
For 0≤n≤365, this is 1−∏ from k=0 to n−1 (1−k/365).
For n≥366, a match is certain in the 365-date model because more people than date categories have been assigned.
Worked example 3: six people
For six fictional people, the no-match product is
(365/365)(364/365)(363/365)(362/365)(361/365)(360/365).
Subtract it from one to obtain the probability of at least one shared birthday. The calculation needs all six factors; multiplying 1/365 by the number of people would not correctly model the event.
Twenty-three participants give 253 pairs
The number of unordered pairs of people is C(n,2)=n(n−1)/2.
At n=23, this becomes 23×22/2=253.
Every pair is a possible matching pair. These pairs are not independent events, so we cannot simply calculate 1−(364/365)^253 as an exact formula.
Worked example 4: why pair count grows quickly
For n=10, there are 10×9/2=45 pairs.
For n=20, there are 20×19/2=190 pairs.
For n=30, there are 30×29/2=435 pairs. Increasing group size threefold from ten to thirty increases the number of pairs by nearly tenfold.
The threshold calculation
Under the exact product formula, a group of 22 has collision probability just under one-half.
For 23, the value becomes about 0.5073, just over one-half.
Therefore twenty-three is the smallest group size meeting the more-likely-than-not condition under the specified 365-day independent uniform model.
Important group sizes
| Number of people | Probability of at least one match |
|---|---|
| 2 | about 0.27% |
| 10 | about 11.69% |
| 20 | about 41.14% |
| 22 | about 47.57% |
| 23 | about 50.73% |
| 30 | about 70.63% |
| 40 | about 89.12% |
| 50 | about 97.04% |
The percentages are rounded model probabilities, not statistics about any real educational group.
Why one match is likely but not guaranteed at twenty-three
A probability of about 50.73% is only slightly above one-half.
Almost half of hypothetical independent 23-person groups still have all different birthday dates.
“More likely than not” is therefore much weaker than “certain.” It does not justify claiming every imaginary classroom of twenty-three must contain a match.
The birthday paradox has a separate certainty threshold
With 365 possible birthday categories, 366 assigned people force a repetition.
That is the pigeonhole principle, a deterministic argument.
At n=23, the collision probability is above one-half but not one. At n=366, a collision is mathematically unavoidable under the date-label model. These are entirely different threshold questions.
Worked example 5: why 365 people do not guarantee a match
Imagine one person on each of the 365 available date labels.
All birthdays are different, so a 365-person group can avoid a match.
Add one more person and there are now 366 assignments into 365 date categories. At least two people must share a date.
One named person’s birthday is another problem
Suppose one particular person’s date is fixed, and we ask whether anyone else in a group matches that person’s birthday.
Under the model, each of the other n−1 people independently avoids that date with probability 364/365.
Thus the chance of a match with the named person is 1−(364/365)^(n−1).
Worked example 6: a named person’s chance at twenty-three
For a group of 23 including one designated person, there are 22 other birthdays.
The chance one or more match the designated date is 1−(364/365)^22≈5.86%.
That is dramatically below 50.73%, because matches among the other twenty-two people do not count for this narrower event.
How event definitions change the denominator
“Any pair matches” considers 253 possible pairs in a 23-person group.
“Someone matches one specified person” considers only 22 comparisons.
The first event includes the second, but also many other ways to create a match. The probability calculations must reflect which event was actually requested.
Expected number of matching pairs
Let Iᵢⱼ be one when people i and j share a birthday and zero otherwise.
There are C(n,2) such pairs. Under the uniform independent model, E[Iᵢⱼ]=1/365.
By linearity of expectation, the expected total number of matching pairs is C(n,2)/365, even though the pair-match indicator variables are not generally mutually independent.
Worked example 7: expected pairs at twenty-three
With n=23, there are 253 unordered pairs.
Expected number of matching pairs is 253/365≈0.69315.
That is not the probability of at least one match. The collision probability is about 0.5073. An expected count can differ from the chance of a nonzero count because some groups contain multiple matching pairs.
Why the expected count can exceed the collision probability
When a collision event occurs, the number of matching pairs could be one, two, three or more.
The expectation weights those possible counts. The event probability merely asks whether the count is at least one.
Therefore these quantities should not be used interchangeably, even when their values happen to be close in some small-group examples.
Worked example 8: one triple produces three pairs
Suppose exactly three fictional people share the same date.
Those people create pairs AB, AC and BC, for three matching pairs.
Yet the yes-or-no event “a collision occurred” is counted just once for the entire group. This explains why pair expectations and collision probabilities measure different things.
A useful approximation for the no-match probability
For modest n relative to 365, use ln(1−u)≈−u when u is small.
Taking logarithms of the no-match product gives approximately
ln P(no match)≈−[0+1+⋯+(n−1)]/365=−n(n−1)/(730).
Exponentiating gives P(no match)≈exp[−n(n−1)/730].
Worked example 9: approximate the twenty-three-person result
For n=23, n(n−1)=23×22=506.
Approximate P(no match)≈e^(−506/730)=e^(−253/365)≈0.5000.
Thus the approximate collision probability is about one-half. The exact product gives around 0.5073, making the approximation informative but not identical.
The approximate fifty-percent threshold
Set e^[−n(n−1)/(2D)]≈1/2 for D equally likely categories.
Taking logarithms gives n(n−1)≈2D ln2.
For D=365, the right side is about 506.0, suggesting n around 23. This reveals the scaling law behind the surprise.
The square-root rule
For large D and n much smaller than D, approximate n(n−1) by n².
The fifty-percent threshold is then near n≈√(2D ln2).
The group size needed for a likely collision grows roughly with the square root of the number of categories, not linearly with D.
Worked example 10: hypothetical thousand categories
Suppose there are D=1,000 equally likely independent labels.
The approximate half-collision threshold is √(2×1,000×ln2)≈37.23.
A group around 38 may therefore cross one-half under the approximate model. An exact decision at a boundary should use the product rather than relying only on the approximation.
Why uniformity matters
The simple product formula assumes every date label has probability 1/365 for every independent person.
Actual birthday frequencies need not be equal across calendar dates.
Unequal category probabilities can change the collision probability. A classroom exercise should state the uniform assumption rather than treating it as an observed demographic fact.
General unequal birthday probabilities
Suppose date label i occurs with probability pᵢ, with all pᵢ nonnegative and summing to one.
For two independently sampled people, the chance of a match is Σ pᵢ², not necessarily 1/D.
Under uniform probabilities pᵢ=1/D, this sum becomes D×(1/D)²=1/D.
Worked example 11: unequal two-category labels
Imagine a fictional two-category system with probabilities 0.8 and 0.2.
The chance that two independent assignments match is 0.8²+0.2²=0.64+0.04=0.68.
Under two equally likely categories, the matching chance would be 0.5. The unequal distribution produces more collisions in this example.
Independence is a separate assumption
Even if every individual marginal birthday were uniformly distributed, group assignments might be dependent.
For example, a group consisting of twins or other related individuals may not behave like independent draws from a 365-date distribution.
The probability product cannot be assumed valid solely because there are 365 labels.
Worked example 12: a deliberately dependent case
Imagine a fictional rule that assigns person B exactly the same date label as person A.
Each individual may still have a uniform marginal distribution, but their birthdays match with probability one, not 1/365.
The difference is caused by dependence between assignments. Uniformity alone does not prove independence.
Simulating collisions is not the same as proving the exact formula
A program can repeatedly generate 23 independent uniform birthday labels and record whether a collision occurred.
Across many repetitions, the simulated proportion should usually be near the exact probability, with random sampling variation.
But simulation results need not equal 0.5073 exactly. The mathematical product formula provides the theoretical reference under the model.
Worked example 13: ten simulated groups
Suppose a toy simulation runs only ten fictional groups of twenty-three.
It might observe matches in four groups, or six, or another count. Those proportions would be 40%,60%, and so on.
Such small simulated batches do not refute the exact 50.73% probability. They illustrate random variation around the model’s long-run event frequency.
A probability is not a personal forecast
A 50.73% group collision chance does not determine whether a specific student shares a birthday with another.
Before observing the group, the theorem quantifies the event probability under hypothetical random sampling.
After inspecting actual dates, the matching condition can be checked directly. The probability calculation is not a substitute for factual observation.
Collisions appear in computer identifiers too
The same mathematical structure arises when independent values are mapped uniformly into a finite set of possible codes.
The birthday bound helps estimate how many generated codes may be needed before collisions become plausible.
But a real identifier system may use deterministic rules, nonuniform distributions, checks or collision avoidance. The birthday model alone does not certify security or reliability.
Worked example 14: four equally likely labels
Let D=4 and n=3.
The no-collision probability is (4/4)(3/4)(2/4)=3/8.
Therefore the collision probability is 5/8=62.5%. This small toy example can be fully enumerated without large factorials.
A complete enumeration for three labels
For three independent draws from three equally likely categories, there are 3³=27 ordered outcomes.
No-match outcomes select three distinct labels in 3!=6 ways.
Thus P(no match)=6/27=2/9 and P(at least one match)=7/9≈77.78%. Enumeration verifies the product 1×2/3×1/3=2/9.
Worked example 15: use the small enumeration
Let labels be A,B,C and draw three times independently.
No collisions occur only in permutations ABC,ACB,BAC,BCA,CAB,CBA.
Six of 27 outcomes avoid matches. The remaining 21 contain at least one repeated label, so probability is 21/27=7/9.
Counting without replacement changes everything
Suppose birthdays were assigned without replacement from a deck containing one card for each of 365 date labels.
Then the first 365 assignments would necessarily be distinct.
That is not the usual birthday model, where multiple people may independently receive the same label. The sampling mechanism must be stated before selecting a formula.
Worked example 16: replacement versus no replacement
Take four distinct labels and draw two labels uniformly.
With replacement, the second can match the first, giving collision probability 1/4.
Without replacement, the second must differ from the first, giving collision probability zero. Same labels, different sampling rule.
The complement should be explicitly stated
Write M for “at least one birthday match.”
Its complement Mᶜ is “all birthday labels are distinct.”
The two events are mutually exclusive and exhaustive. Therefore P(M)=1−P(Mᶜ). This logical structure is the foundation of the computation.
Why adding pair probabilities is not exact
With n people, a naive sum over C(n,2) pairs gives expected matching-pair count C(n,2)/365.
But the events “A matches B” and “A matches C” can occur simultaneously.
Adding their probabilities double-counts those overlaps. Inclusion–exclusion could correct the overlaps, but the complement product is much simpler here.
Worked example 17: a too-large naive sum
For n=50, the number of pairs is 50×49/2=1,225.
Summing pair-match probabilities gives 1,225/365≈3.356.
That cannot be an event probability because it exceeds one. It is the expected number of matching pairs, demonstrating why the naive probability interpretation fails.
Different collision questions demand different formulas
“At least one pair” differs from “exactly one pair,” “at least one triple,” and “one named person’s birthday is matched.”
Each event has a different set of favourable outcomes.
The birthday product used here solves only the first question. More restrictive collision patterns need additional combinatorial counting or generating functions.
Worked example 18: exactly one match among three
For three independent uniform draws from D date labels, exactly one matching pair means two identical labels and a third different label.
Choose which two people match in C(3,2)=3 ways. Choose the shared label in D ways and the different label in D−1 ways.
Probability is 3D(D−1)/D³=3(D−1)/D². This excludes the all-three-same case.
A triple also creates pair collisions
For three people, the all-three-same event has probability 1/D² under uniform independent sampling.
It contains three matching pairs, so it belongs to the “at least one pair” event but not the “exactly one matching pair” event.
Distinguishing such overlapping events prevents careless counting.
Worked example 19: verify the three-person decomposition
For n=3 and D uniform labels:
P(exactly one matching pair)=3(D−1)/D². P(all three match)=1/D².
Their sum is (3D−2)/D², which equals 1−D(D−1)(D−2)/D³ after simplification. Both approaches compute the same at-least-one-pair probability.
The actual birthday distribution may be seasonal
In real populations, birth rates can vary across days or seasons, and February 29 exists only in leap years.
These effects change the true distribution of birthday labels, especially when comparing groups with particular age or demographic characteristics.
The common 23-person result remains a valuable mathematical teaching example, but its exact 50.73% value belongs to the uniform 365-label model.
A leap-day model needs explicit probabilities
Simply changing D from 365 to 366 and assuming all 366 labels equally likely is one alternative mathematical model.
It does not automatically describe actual birthdays, because February 29 occurs less often across years than ordinary dates.
To calculate a real-population probability, one would need valid date probabilities and information about dependence in the sampled group.
Worked example 20: uniform 366-day thought experiment
Suppose, purely hypothetically, every one of 366 labels is equally likely.
The two-person matching probability becomes 1/366 instead of 1/365.
The exact group formula uses factors (366−k)/366. That is a changed model, not a factual claim about actual birth-date frequencies.
Birth-year equality is a different condition
Two people sharing a birthday usually means they have the same month and day, regardless of year.
If the question instead requires identical year, month and day, the number and distribution of categories depend on the years in scope.
The 365-day collision model is not an answer to a full-date-of-birth identity question.
A collision probability increases with group size
Adding a person cannot remove an existing match among the earlier people.
Therefore the event “at least one collision” grows as the group expands, so its probability is nondecreasing in n under the stated joint sampling model.
It begins at zero for n=0 or n=1, then rises toward certainty by n=366 in the 365-label setting.
Worked example 21: first two group sizes
For n=1, no pair exists, so P(match)=0.
For n=2, P(match)=1/365.
For n=3, the probability is larger, because the event includes all previous two-person matches plus possible new matches involving the third person.
The probability is not linear in group size
Doubling the number of people generally creates more than twice as many pair comparisons.
The pair count n(n−1)/2 grows approximately quadratically for large n.
That growth helps explain why collision probabilities climb rapidly once group size approaches the square-root scale of the number of possible labels.
Worked example 22: compare pair counts
At n=15, there are 105 pairs.
At n=30, there are 435 pairs.
Doubling the group size increases potential pairs by more than fourfold in this comparison. The matching events overlap, so probabilities do not scale by the same exact factor, but the count reveals the pressure toward collisions.
A probability of ninety percent is still not certainty
At n=40, the usual model gives a collision probability of about 89.12%, just under 90%.
At n=50, it is around 97.04%.
Even at 50, the event is not logically guaranteed. All 50 people could have distinct birthday labels, though that arrangement is unlikely under the uniform independent model.
The distinction from pigeonhole guarantees
Our earlier Pigeonhole Principle and Guaranteed Matches guide asks what must happen regardless of how categories are assigned.
The birthday paradox instead asks how likely an event is under a stochastic assignment model.
These are complementary questions. One uses counting inevitability; the other uses probabilities over possible assignments.
A useful arithmetic implementation
To calculate no-match probability for n people, start q=1.
For k=0,1,…,n−1, update q by multiplying (365−k)/365. Finally return 1−q.
This avoids calculating gigantic factorials directly. For n≤365, each factor is between zero and one, making the logic easy to audit.
Worked example 23: calculate through three steps
Start q₀=1.
After first person, q₁=1×365/365=1. After second, q₂=364/365. After third, q₃=(364/365)(363/365).
Subtract from one to obtain the three-person collision probability. This stepwise process is exactly the product formula.
Floating-point calculations still need care
For large factorials, direct numerical evaluation can overflow or lose precision.
Multiplying factors progressively, or adding their logarithms for very large category sets, is often more stable.
The numerical method does not change the underlying combinatorial proof. It only affects how reliably a machine evaluates the formula.
Common error 1: confuse one specified birthday with any pair
A named person’s matching probability in a 23-person group is about 5.86%.
The chance of some match anywhere in the group is about 50.73%.
These are different events. The repair is to write the sample-space question in words before choosing the formula.
Common error 2: add pair probabilities directly
A group of many people contains overlapping possible matching events.
Adding pairwise probabilities may exceed one and actually computes the expected number of matching pairs.
Use the complement product for the probability of at least one collision.
Common error 3: claim twenty-three people guarantee a match
A probability just over 50% means a match is more likely than not.
It does not force one. The 365-category model guarantees a match only with 366 assigned people.
Mathematical certainty and high probability are distinct conclusions.
Common error 4: forget independence
The simple product uses independent birthday assignments.
If a group has dependent date labels, the conditional factors may differ.
A real population requires an appropriate data-generating model; writing 1/365 everywhere without justification can overstate precision.
Common error 5: confuse 365 and 366 labels
The familiar simplified result uses 365 equally likely dates and ignores leap day.
A full model including leap-day birthdays needs its own category probabilities.
State the date-label convention rather than treating the number of categories as invisible.
Common error 6: count pairs as n squared
Each unordered pair of distinct people is counted twice by an ordered count n(n−1).
Divide by two to obtain C(n,2).
At n=23, the correct unordered count is 253, not 506 or 529.
Common error 7: treat an expected pair count as the event probability
At n=23, expected matching pairs are approximately 0.693.
The actual probability of at least one pair is approximately 0.507.
Multiple pairs can occur in one group, so the expectation and the yes-or-no event probability are not the same.
Common error 8: infer real classroom dates from model probabilities
The mathematical calculation does not tell us whether any specific actual group has a shared birthday.
Actual birthday matches require actual observations, with appropriate respect for personal information.
Our fictional example never uses or reveals real pupil data.
Primary readiness: months and simple matching
Primary learners can classify imaginary dates by month and discover that thirteen people guarantee at least two in the same month.
This deterministic puzzle precedes the more advanced birthday probability calculation.
The difference between one named match and any pair can be explained with small coloured-card examples.
Lower Secondary: multiplication and complementary probability
Students can compute no-match probabilities for two, three and four uniform labels.
They can then extend the same logic to 365 categories and use a calculator to evaluate successive factors.
Understanding the complement, denominator and independence assumptions matters more than memorising the number 23.
Advanced probability and syllabus boundaries
The birthday paradox connects combinations, sampling, collision counting and probability approximation.
Consult the SEAB 2027 G3 Mathematics and Additional Mathematics syllabus directory for the learner’s school scope. The large-group approximation and information-system parallels are enrichment rather than claims of compulsory examination content.
A twenty-minute family diagnostic
Minutes 1–5: Ask whether thirteen people must share a month, and explain why.
Minutes 6–10: Calculate P(no match) for three people using a toy system of four equally likely date labels.
Minutes 11–15: Write the full 365-label no-match product and compute the complementary probability for n=23.
Minutes 16–20: Compare the 50.73% any-pair result with the 5.86% named-person result and the 366-person certainty threshold.
Seven days of independent practice
Day one practises pigeonhole certainty. Day two counts unordered pairs.
Day three uses complementary events and multiplication. Day four calculates no-match products for small D.
Day five evaluates the 365-date model. Day six derives the expected number of matching pairs. Day seven analyses nonuniform labels and distinguishes simulation from an exact model probability.
Practice questions with checked answers
A. A group of two, uniform 365 labels: match probability? Answer: 1/365.
B. Number of unordered pairs among 23 people? Answer: 253.
C. Any matching birthday among 23 people under the standard model? Answer: about 50.73%.
D. Named person’s birthday matched among 23 people? Answer: about 5.86%.
E. Expected matching pairs among 23? Answer: 253/365≈0.69315.
F. Smallest group guaranteeing a birthday match among 365 labels? Answer: 366.
G. Three independent uniform labels from three categories: any collision? Answer: 7/9.
H. Does 97% chance mean a match is certain? Answer: no.
Frequently asked questions
Why do just 23 people produce a 50% chance of shared birthdays?
They create 253 different possible pairs, greatly increasing opportunities for a collision under independent uniform dates.
Is the result exactly 50%?
No. Under the standard 365-day model it is approximately 50.73% at 23 people.
Does the birthday paradox rely on a contradiction?
No. The answer feels surprising, but it follows consistently from complementary probability and pair counting.
Are real birthdays equally likely?
Not necessarily. Equal date probabilities are a simplifying assumption, and real distributions can differ.
Is a match guaranteed with 23 people?
No. It is merely more likely than not in the specified model.
Why do 366 people guarantee a match?
There are only 365 date labels in that model, so pigeonhole counting forces a repeated label.
What if I ask for someone matching my particular birthday?
That has a smaller probability because matches among other people are excluded.
What is the connection to computer identifiers?
Independent random identifiers from a finite code space can collide according to similar pair-counting mathematics, under the correct distribution assumptions.
Are these actual Bukit Timah school birthday statistics?
No. Every group and birthday assignment is a hypothetical mathematical construction.
How can parents recognise understanding?
The student states the event, derives the no-match complement, counts pairs correctly and distinguishes chance from a logical guarantee.
Official context and related Mathematics in Bukit Timah
See the SEAB 2027 G3 Mathematics syllabus directory for school-level probability foundations. All group sizes here are model examples, not actual local class sizes or birth-date data.
Continue with Pigeonhole Principle and Guaranteed Matches, Probability and Independent Events, Expected Values and Probability Distributions and Combinations and Pascal’s Triangle.
The immutable Clementi Secondary 1 Mathematics tutorial reference anchors explanations from first principles, checked arithmetic, precise conditions and independent transfer.
The lasting habit is to define the matching event, calculate its complement, count the potential pairs, check the sampling assumptions and never mistake a probability threshold for certainty.
Companion Mathematics in Bukit Timah articles from this November batch: Monty Hall Problem, Switching Doors and Conditional Probability · Spherical Geometry, Great Circles and Shortest Surface Distances · Mathematical Invariants, Parity Puzzles and Impossible Moves. These four original articles distinguish mathematical probability from certainty, conditioned decisions from naive guessing, surface geometry from flat distance, and structural invariants from trial-and-error searches.