In 1889, Francis Galton described a board of pegs that converts pure randomness into a perfect bell curve. Every time.
scroll to begin
Francis Galton called it a "quincunx" — a triangular grid of pegs with a slot at the top and collection bins at the bottom. Drop a ball in, and gravity does the rest.
At each row of pegs, the ball bounces either left or right with equal probability — a coin flip. With 12 rows, the ball makes 12 independent random choices on its way down.
The ball's final bin is the number of times it bounced right. Every path through the board is a sum of random events.
There are many more paths leading to the center than to the extremes.
Drop ten more. Each path is different. Already, more balls land near the center than at the edges — there are C(12, 6) = 924 paths to the center bin, but only 1 path to each extreme.
The shape sharpens. The center bins fill fastest, and the distribution tapers symmetrically toward the edges. A bell is forming.
Unmistakable. A smooth, symmetric bell curve — entirely from random bounces. No ball was aimed. No bin was favored. The shape came from arithmetic alone.
The green curve is the theoretical prediction — the binomial distribution for 12 fair coin flips. No fitting, no free parameters.
This is the central limit theorem made physical.
Every individual path changes. The shape doesn't. This is Galton's insight: random micro-behavior produces predictable macro-patterns.
Each ball's final position is a sum: right + left + right + right + left + … — twelve independent coin flips. The central limit theorem guarantees that such sums converge to a normal distribution, regardless of what the individual terms look like.
The Galton board doesn't produce the bell curve because of some special property of pegs. It produces it because addition is happening.
Each line tracks the running average landing bin as balls accumulate. They all converge to the theoretical mean.
20 independent simulations, 500 balls each, 12 rows. Dashed line: theoretical mean (6.0).
With 12 rows and a fair bounce, each ball follows a Binomial(12, 0.5) distribution over 13 bins. The probability of landing in bin k is C(12, k) / 212. The center bin (k = 6) has probability 22.6%. The extreme bins (k = 0 or k = 12) have probability 0.024%.
As the number of rows grows, this binomial converges to a Gaussian with mean N/2 and standard deviation √N / 2.
The histogram matches the theoretical distribution almost exactly. Individual paths are random; the aggregate is not.
10,000 simulated balls, 12 rows. Green curve: Binomial(12, 0.5) expected counts.
Measurement errors are sums of many small disturbances. Human heights are influenced by many genes. Exam scores aggregate many question-level performances. Stock returns compound many individual trades.
In each case, many small independent effects are added together. Galton's board makes this universal pattern visible: the bell curve is not a property of any one domain — it is a property of addition itself.
Adjust the number of rows and the bounce bias. With more rows, the curve smooths out. With biased bounces, the center shifts.