The Birthday Paradox Picked a Fight With the Founding Fathers and lost

There is a party trick that mathematicians love. Get 23 people in a room, and the odds are better than even that two of them share a birthday. Get 56 people together, and a shared birthday is close to a sure thing. Not likely. Not probable. Close to certain, at very nearly 99 percent.

So here is a question with a clean setup. The Declaration of Independence was signed by 56 men. If the math is that confident, at least one pair of those signers should share a birthday, right?

I went looking for that pair. The answer surprised me, and the reason it surprised me is a small lesson about how probability actually works once it stops being a party trick and has to survive contact with real people and real records.

The reason the birthday problem feels wrong at first is that most people quietly answer a different question. They imagine walking into a room and asking, "Does anyone here share my birthday?" That version really is unlikely with a small group. You are comparing yourself against everyone else, and that is only 22 comparisons in a room of 23.

But the real question does not care about you specifically. It asks whether any two people match, and the number of possible pairings grows fast. In a room of 23 people there are 253 different pairs who could match. In a room of 56, there are 1,540 pairs. Each pair is its own little lottery ticket, and once you are holding well over a thousand tickets, the chance that none of them wins gets very small.

The cleanest way to run the numbers is to flip the question around and ask for the odds that everyone is different. The first person can have any birthday. The second has to dodge one date, so 364 out of 365. The third has to dodge two, the fourth has to dodge three, and so on. You multiply all those shrinking fractions together, and by the time you reach 23 people the product drops below one half. That crossover point is what makes 23 the famous number. Push on to 56 people and the odds that everyone stays unique fall to a little over one percent, which puts the chance of a shared birthday at about 98.8 percent.

That figure assumes a tidy world. Every birthday equally likely. Every date known. Nobody missing from the records. Hold onto that, because it is exactly where the founders wandered off the map.

Before we go looking at the actual records, it helps to separate two very different kinds of certainty, because the birthday paradox is really a wrestling match between them.

The first kind is a guarantee. There is a rule in mathematics called the pigeonhole principle, and it is almost insultingly simple to state. If you have more pigeons than pigeonholes, at least one hole has to hold two pigeons. Ten letters going into nine mailboxes means some mailbox gets two. There is no probability here, no "likely," no wiggle room. It is a flat certainty.

Turn that on birthdays. There are 366 possible birthdays once you count February 29. So the moment you put 367 people in a room, a shared birthday stops being probable and becomes unavoidable. You have run out of unique dates to hand out, and the 367th person is forced to double up with someone. That is the hard ceiling. No collection of 367 or more people can ever dodge a shared birthday, no matter how the dates fall.

The second kind of certainty is the soft, probabilistic one we have been using. It does not force anything. It just tells you the odds. And the whole reason the birthday paradox feels like a paradox is the enormous gap between these two numbers. The pigeonhole guarantee needs 367 people. The coin-flip odds need only 23. In the wide territory between those two figures, a shared birthday is likely, sometimes overwhelmingly likely, but never promised.

The whole drama lives in the gap: 23 for a coin flip, 367 for a guarantee. The 56 signers sit deep in the probabilistic zone.

That gap is the entire drama of the founders. Fifty six signers is a long way from 367. Nobody was ever forced. The signers lived deep inside the probabilistic zone, the stretch where the math can whisper "almost certainly" and still be talking about odds rather than laws. And odds, unlike the pigeonhole principle, are allowed to lose.

The National Archives keeps a factsheet on the signers with birth dates, birthplaces, occupations, and ages. It is the closest thing to an official ledger, and it is where I started.

Going down the list one date at a time, you find some close calls. John Adams was born on October 30. William Paca was born on October 31, one day off. Benjamin Franklin came into the world on January 17, and William Whipple on January 14, three days apart. Tantalizing, but a miss is a miss. A shared birthday means the same month and the same day, and after checking every name against every other name, the ledger gives up nothing. Not one confirmed pair.

The closest calls in the ledger. A miss is a miss - and after every name against every other name, not one confirmed pair.

For an event the math rated at almost 99 percent, zero matches feels like watching a coin land on its edge.

The clean calculation collapses because of a boring, human problem. The eighteenth century did not keep birthday records the way we do.

About nine of the signers have no verified birth date at all. Their entries carry a small "circa" and a year, nothing more. Button Gwinnett, John Hart, Thomas Lynch Jr., John Morton, James Smith, Thomas Stone, George Taylor, Matthew Thornton, George Wythe. For these men, historians can point to a rough year and sometimes not even that. George Walton's birth is placed somewhere in a ten year window depending on which source you trust.

You cannot ask whether two dates match when you do not have the dates. So the real pool of comparable signers is not 56. It is closer to 46 or 47. And that one change matters more than it looks. Run the birthday math on 47 people instead of 56 and the odds of a shared birthday slip from near certainty down to roughly 95 percent. Still high. Still the way to bet. But 95 percent leaves about a 4.5 percent escape hatch, and reality happened to walk through it.

That is the quiet point worth keeping. A 4.5 percent event is not a broken calculation or a paradox that failed. It is simply the thing that happens about one time in twenty-two. Someone has to be the one in twenty-two. This time it was a room full of revolutionaries.

If you want to see why the number lands near 5 percent without wading through a chain of shrinking fractions, there is a cleaner path, and it has a name. Statisticians call it the Poisson approximation, and it is the standard tool for a very specific situation: when you have a large number of chances for something to happen, each chance is individually tiny, and you care about how many times the rare thing occurs across all of them. Shared birthdays fit that description almost perfectly. There are over a thousand pairs of signers, and any single pair matching is a long shot at 1 in 365. That is exactly the shape the Poisson approximation was built for.

The method starts by counting the chances, not the people. With 47 signers who have known dates, the number of possible pairings is 47 times 46 divided by 2, which comes to 1,081 pairs. Each pair has a 1 in 365 chance of matching. Multiply the pairs by that chance and you get the expected number of shared birthdays: 1,081 divided by 365, which is right around three. So on average you would expect about three matching pairs in a group this size. Statisticians label that expected count with the Greek letter lambda, and here lambda is three.

Count the chances, not the people. Expected matches is lambda; the odds of zero is e to the negative lambda.

The reason the Poisson approximation is worth reaching for is what it does next. Once you know lambda, it hands you the full spread of outcomes, not just the average. It can tell you the odds of exactly zero matches, exactly one, exactly two, and so on. The formula for landing on any particular count involves lambda, that count, and the constant e, which is the familiar 2.718 that turns up all over mathematics. For our purposes only the zero case matters, and there the formula collapses to something you can almost do in your head: the probability of zero events is e raised to the power of negative lambda. Plug in lambda of three and you get e to the negative three, which works out to about 0.05, or 5 percent.

There is a quiet reason to trust this shortcut rather than treat it as a party trick. The Poisson approximation is not a rough guess pulled from nowhere. It is a well understood limit that the exact pair-by-pair probability converges toward whenever the individual events are rare and roughly independent, which is precisely the regime we are in. That is why it holds up here. The pairs are not perfectly independent, since the same person sits in many pairs at once, but with collisions this rare the dependence is weak enough that the approximation barely feels it.

Two completely different methods, both hovering around one chance in twenty. The number is real, not an artifact of either.

That is worth pausing on. Two completely different methods, the long exact multiplication and this named Poisson estimate, both point to nearly the same answer: the exact calculation gives about 4.5 percent, the Poisson shortcut about 5 percent, both hovering around one chance in twenty that a group this size produces no shared birthday at all. When a quick approximation and a grinding exact calculation land that close together, you can be fairly sure the number is real and not an artifact of either method. The historical record did not defy the math. It landed on the small but perfectly ordinary tail the math predicted.

There is a second wrinkle underneath the first. Even the "known" dates come with fine print. Britain and its colonies switched from the Julian to the Gregorian calendar in 1752, sliding every date forward by eleven days. A man born before the switch might have one date in the parish book and a different one on his own tombstone, because he moved his birthday to match the new calendar. Samuel Huntington is the perfect headache here. Different records place his birth on July 3, July 5, and July 16 of 1731, and at least one of those gaps is the calendar shift showing up in the paperwork. When the raw data is this soft, the crisp 98.8 percent was always promising more precision than the historical record can deliver.

Which brings us to the one coincidence the records do seem to like, even if they cannot fully agree on it.

By most accounts, Samuel Huntington of Connecticut was born on July 3, 1731. One day before the date the whole country now marks with fireworks. No signer was born on the Fourth itself, at least not in any record we can trust. But the man came within twenty four hours of sharing a birthday with the nation he helped create, and then spent his life signing the documents that gave the date its meaning. The Declaration, the Articles of Confederation, and a term as President of the Continental Congress.

Three of the first five presidents did manage to die on the Fourth. John Adams and Thomas Jefferson both let go on July 4, 1826, the fiftieth anniversary, within hours of each other. James Monroe followed exactly five years later. That cluster is its own improbable story for another day. But it is worth noting that the calendar seems fonder of collecting founders on the way out than of stamping them with matching birthdays on the way in.

The birthday paradox is usually sold as proof that human intuition is bad at probability, and it is. But the signers add a second lesson that the classroom version leaves out.

The math is only ever as good as the world you feed it. Assume clean, complete, equally likely data and you get a clean, confident 98.8 percent. Feed it a real archive with missing years, smudged town records, and a calendar that jumped eleven days in the middle of everyone's lifetime, and the confident number turns into a range, and the range leaves room for exactly the empty result we found.

“

A probability is a statement about a model, and the model is never the whole world.

There is a reason none of this violates any law of mathematics. The one iron law in play, the pigeonhole principle, only bites at 367 people, and the signers never came close. Everything below that ceiling was a matter of odds, and odds are permitted to surprise you. Had there been 367 founders, a shared birthday would have been carved in stone. With 56, it was only ever a good bet, and good bets lose about as often as their probabilities say they will.

So the founders did not break the birthday paradox. They did something more useful. They reminded us that a probability is a statement about a model, and the model is never the whole world. The likely thing is still the likely thing. It just does not always show up when you go looking for it, and July 4th turned out to be one of those times.