BM · 08 Beauty of Math

The Galton Board: Where the Bell Curve Comes From

The Galton Board: Where the Bell Curve Comes From

Drop balls through rows of pins: at every pin, each ball goes left or right like a coin flip. No ball aims for the middle, yet thousands of them pile up into the same shape every time: a bell. We follow one ball, count the paths with Pascal's triangle, watch the heap wobble and settle as more balls fall, add more rows until the heap becomes de Moivre's smooth curve, tilt the board, roll dice to meet the central limit theorem, see why the bell turns up in nature, follow Galton's own puzzle to regression to the mean, and find where the bell curve breaks.

Chapters

Cheat sheet

Sources

Transcript

Drop a ball onto a board of pins. At every pin, it bounces left or right, like the flip of a coin. Drop another, and another. No ball aims for the middle. Each one is pure chance.

And yet, as thousands pile up, the same shape appears every time: a bell. By the end of this video, you'll know why the middle always wins, how a triangle of numbers predicts the heap, why the same bell shows up in nature, and where it breaks.

This is Just Known. From unknown, to known. Beauty of Math, episode eight: the Galton board, where the bell curve comes from. In the eighteen seventies, Francis Galton, a half-cousin of Charles Darwin, had a board built with rows of pins and a funnel at the top. He called it the quincunx, after the way its pins were staggered, like the five dots on a die, repeated.

Lead shot poured in at the top, bounced down through the pins, and collected in slots at the bottom. He wanted a machine that showed, in a few seconds, a law he saw all through nature. Follow one ball. Twelve rows of pins, so twelve bounces. At each pin, it goes left or right, fifty-fifty.

Its whole trip is a string of twelve lefts and rights. Where it lands depends only on how many times it went right. None, and it ends at the far left. All twelve, the far right. Six, right in the middle.

The order doesn't matter, only the count. So how many trips end in each slot? Count them, row by row. There's one way to reach the first pin. Below it, one way left, and one way right.

On the next row, the middle can be reached two ways: left then right, or right then left. Every count is the sum of the two above it. This is Pascal's triangle. At the bottom, after twelve rows, it reads one, twelve, sixty-six, two hundred twenty, all the way up to nine hundred twenty-four in the middle. There are four thousand ninety-six trips in all, and only one reaches each edge.

That's the whole secret. Every trip is equally likely, but nine hundred twenty-four of them end in the middle slot. So a ball is nine hundred twenty-four times as likely to land in the middle than at the far left. It's like flipping twelve coins: six heads can happen in nine hundred twenty-four ways, twelve heads in just one. Nobody steers the balls.

The middle simply has the most ways in. Divide each count by four thousand ninety-six, and you get the chance of each slot: almost twenty-three percent for the middle, and about one in four thousand for each edge. Now drop a thousand balls. The heap follows the prediction, but not exactly: chance makes it wobble. Drop ten times as many, and the wobble shrinks, though only by about three times.

With a hundred thousand balls, the heap sits within a fraction of a percent of the prediction. Add more rows, and the heap changes shape. With four rows, it's a lumpy staircase. With twelve, a rough bell. With fifty, or two hundred, a smooth curve.

And it grows wider, but slowly: with four times as many rows, the heap is only twice as wide. Its spread grows with the square root of the number of rows. So compared to the whole board, the heap gets narrower and narrower. That smooth curve has a formula. In seventeen thirty-three, Abraham de Moivre showed that for many coin flips, the counts follow a curve shaped like e to the minus x squared, over two.

Today, we call it the normal curve, or the bell curve. It's set by just two numbers: where its middle is, and how wide it is. About sixty-eight percent of everything lands within one width of the middle, and about ninety-five percent within two. For two hundred rows, the board and the formula agree to about a tenth of a percent. What if the board is tilted, so every pin sends a ball right sixty percent of the time?

The heap slides over: its middle moves from six to about seven. But its shape is still a bell, just shifted, and with more rows it gets even closer to one. The coin doesn't have to be fair. It just has to be flipped many times. Here's the big idea.

Each ball's final slot is a sum: twelve small steps, each one zero or one. And sums of many small, independent pieces pile up into a bell. Roll one die, and every number is equally likely: flat. Add two dice, and you get a triangle, with seven on top. Three dice: already curved.

Ten dice: a bell. In eighteen ten, Pierre-Simon Laplace showed the general rule, now called the central limit theorem: add up enough independent random pieces, of almost any shape, and the total follows the bell curve. That's why the bell turns up in nature. A person's height comes from many genes and many small influences adding up, so the heights of grown men, or of grown women, are roughly bell-shaped. Measurement errors add up from many tiny disturbances, so they follow the bell too.

Carl Friedrich Gauss, and astronomers after him, used the curve to combine their observations. Galton saw it in chest sizes, in heights, in exam marks, and his board showed why: many small causes, added together. Galton had a puzzle. Tall parents tend to have tall children, so why don't heights spread wider and wider with every generation? He built a second board to think it through.

Balls stopped halfway, in a row of slots, and then each slot was released to fall the rest of the way. Each slot spread its balls into a small bell of its own, and together they made a bell wider than the one halfway down. So for heights to stay the same, children can't just scatter around their parents. On average, they must land closer to the middle. And they do: the children of very tall parents are tall, but on average less extreme than their parents.

He called it reversion. Today, we call it regression to the mean. But not everything is a bell. If causes multiply instead of adding, like money earning interest on itself, the heap stretches into a long tail. Incomes and city sizes look more like that.

If one big cause dominates, or the pieces aren't independent, the bell can fail too, and rare events can turn out far less rare than the bell promises. The bell curve is the shape of many small, independent things added up. When that's not the story, it's the wrong curve. You can build your own Galton board in about ten lines. Make thirteen empty slots.

For each of ten thousand balls, start at zero, and for each of twelve rows, add a coin flip: zero or one. Then drop the ball in its slot. Finally, draw a bar for each slot. Run it again, and the bars wobble a little, but the bell stays. The code is in the description.

Let's recap. Every pin is a coin flip, and every trip through the board is equally likely. But many more trips end in the middle, as Pascal's triangle counts. Add more rows and the heap becomes the bell curve. And it's not just about boards: whenever many small, independent pieces add up, the bell appears.

When they multiply, or depend on each other, it often doesn't. Next time: Euler's formula, where spinning around a circle leads to one equation that ties together e, i, pi, one and zero. This is Just Known. From unknown, to known.

Note: every heap, count and curve is computed: the balls are simulated with a seeded random generator, the predictions are exact. Pictures marked "illustration" are schematic.
Narration voice generated with Google Gemini text-to-speech.