WEBVTT

00:00:10.048 --> 00:00:11.933
Imagine a group of people.

00:00:11.933 --> 00:00:14.304
How big do you think the group
would have to be

00:00:14.304 --> 00:00:18.778
before there's more than a 50% chance
that two people in the group

00:00:18.778 --> 00:00:21.218
have the same birthday?

00:00:21.218 --> 00:00:24.187
Assume for the sake of argument
that there are no twins,

00:00:24.187 --> 00:00:26.748
that every birthday is equally likely,

00:00:26.748 --> 00:00:29.977
and ignore leap years.

00:00:29.977 --> 00:00:33.049
Take a moment to think about it.

00:00:33.049 --> 00:00:35.908
The answer may seem surprisingly low.

00:00:35.908 --> 00:00:37.708
In a group of 23 people,

00:00:37.708 --> 00:00:44.669
there's a 50.73% chance that 
two people will share the same birthday.

00:00:44.669 --> 00:00:47.239
But with 365 days in a year,

00:00:47.239 --> 00:00:50.489
how's it possible that you need such
a small group

00:00:50.489 --> 00:00:53.700
to get even odds of a shared birthday?

00:00:53.700 --> 00:00:58.156
Why is our intuition so wrong?

00:00:58.156 --> 00:00:59.498
To figure out the answer,

00:00:59.498 --> 00:01:01.389
let's look at one way a mathematician

00:01:01.389 --> 00:01:05.218
might calculate 
the odds of a birthday match.

00:01:05.218 --> 00:01:09.110
We can use a field of mathematics
known as combinatorics,

00:01:09.110 --> 00:01:14.419
which deals with the likelihoods
of different combinations.

00:01:14.419 --> 00:01:16.950
The first step is to flip the problem.

00:01:16.950 --> 00:01:21.330
Trying to calculate the odds 
of a match directly is challenging

00:01:21.330 --> 00:01:25.229
because there are many ways you
could get a birthday match in a group.

00:01:25.229 --> 00:01:31.389
Instead, it's easier to calculate the odds
that everyone's birthday is different.

00:01:31.389 --> 00:01:32.820
How does that help?

00:01:32.820 --> 00:01:35.741
Either there's a birthday match
in the group, or there isn't,

00:01:35.741 --> 00:01:38.461
so the odds of a match
and the odds of no match

00:01:38.461 --> 00:01:41.860
must add up to 100%.

00:01:41.860 --> 00:01:44.271
That means we can find 
the probability of a match

00:01:44.271 --> 00:01:50.381
by subtracting the probability
of no match from 100.

00:01:50.381 --> 00:01:53.806
To calculate the odds of no match,
start small.

00:01:53.806 --> 00:01:58.281
Calculate the odds that just one pair
of people have different birthdays.

00:01:58.281 --> 00:02:00.632
One day of the year will be
Person A's birthday,

00:02:00.632 --> 00:02:06.022
which leaves only 364 possible birthdays
for Person B.

00:02:06.022 --> 00:02:10.592
The probability of different birthdays
for A and B, or any pair of people,

00:02:10.592 --> 00:02:14.412
is 364 out of 365,

00:02:14.412 --> 00:02:20.514
about 0.997, or 99.7%, pretty high.

00:02:20.514 --> 00:02:22.562
Bring in Person C.

00:02:22.562 --> 00:02:25.793
The probability that she has 
a unique birthday in this small group

00:02:25.793 --> 00:02:29.532
is 363 out of 365

00:02:29.532 --> 00:02:33.964
because there are two birthdates
already accounted for by A and B.

00:02:33.964 --> 00:02:38.582
D's odds will be 362 out of 365,
and so on,

00:02:38.582 --> 00:02:44.474
all the way down to W's odds
of 343 out of 365.

00:02:44.474 --> 00:02:46.385
Multiply all of those terms together,

00:02:46.385 --> 00:02:50.942
and you'll get the probability
that no one shares a birthday.

00:02:50.942 --> 00:02:54.064
This works out to 0.4927,

00:02:54.064 --> 00:03:01.362
so there's a 49.27% chance that no one in
the group of 23 people shares a birthday.

00:03:01.362 --> 00:03:05.955
When we subtract that from 100,
we get a 50.73% chance

00:03:05.955 --> 00:03:08.701
of at least one birthday match,

00:03:08.701 --> 00:03:11.955
better than even odds.

00:03:11.955 --> 00:03:16.144
The key to such a high probability 
of a match in a relatively small group

00:03:16.144 --> 00:03:20.325
is the surprisingly large number
of possible pairs.

00:03:20.325 --> 00:03:26.017
As a group grows, the number of possible 
combinations gets bigger much faster.

00:03:26.017 --> 00:03:29.196
A group of five people 
has ten possible pairs.

00:03:29.196 --> 00:03:32.905
Each of the five people can be paired
with any of the other four.

00:03:32.905 --> 00:03:34.835
Half of those combinations are redundant

00:03:34.835 --> 00:03:39.615
because pairing Person A with Person B
is the same as pairing B with A,

00:03:39.615 --> 00:03:41.685
so we divide by two.

00:03:41.685 --> 00:03:43.045
By the same reasoning,

00:03:43.045 --> 00:03:45.836
a group of ten people has 45 pairs,

00:03:45.836 --> 00:03:49.835
and a group of 23 has 253.

00:03:49.835 --> 00:03:52.905
The number of pairs grows quadratically,

00:03:52.905 --> 00:03:57.665
meaning it's proportional to the square
of the number of people in the group.

00:03:57.665 --> 00:04:00.966
Unfortunately, our brains 
are notoriously bad

00:04:00.966 --> 00:04:04.447
at intuitively grasping 
non-linear functions.

00:04:04.447 --> 00:04:11.235
So it seems improbable at first that 23
people could produce 253 possible pairs.

00:04:11.235 --> 00:04:15.267
Once our brains accept that,
the birthday problem makes more sense.

00:04:15.267 --> 00:04:20.135
Every one of those 253 pairs is a chance
for a birthday match.

00:04:20.135 --> 00:04:22.897
For the same reason,
in a group of 70 people,

00:04:22.897 --> 00:04:26.616
there are 2,415 possible pairs,

00:04:26.616 --> 00:04:33.337
and the probability that two people
have the same birthday is more than 99.9%.

00:04:33.337 --> 00:04:36.707
The birthday problem is just one example
where math can show

00:04:36.707 --> 00:04:38.917
that things that seem impossible,

00:04:38.917 --> 00:04:41.410
like the same person winning
the lottery twice,

00:04:41.410 --> 00:04:44.551
actually aren't unlikely at all.

00:04:44.551 --> 00:04:48.868
Sometimes coincidences aren't
as coincidental as they seem.