Probability & statistics · 03 · Order out of chaos · 8 min
The bell curve
Heights, measurement errors, test scores, blood pressure — wildly different things keep producing the same gentle bell-shaped pile. That's not coincidence. It's the central limit theorem, the closest thing statistics has to magic.
Build the intuition
Why bells appear everywhere
Add up many small, independent influences — thousands of genes nudging height, dozens of tiny errors nudging a measurement — and their sum piles into a bell, almost regardless of what the individual influences look like. Averages of almost anything turn normal. The bell isn't an assumption; it's an emergent law.
68 — 95 — 99.7
One shape means universal rules: about 68% of values sit within 1σ of the mean, 95% within 2σ, 99.7% within 3σ. A blood result “2.5σ high” is rare (≈1%) regardless of what's measured — the σ-ruler works on everything bell-shaped.
Calculus cameo: area is probability
On a continuous curve, probability is area underneath — “chance of a value between a and b” is the integral of the curve from a to b. The 68% rule is literally the area within ±1σ. If you've met integrals, you already own the key concept.
See it move
Widen the ±kσ window and watch the captured area: 68% at 1σ, 95% at 2σ. The shaded area is probability.
A worked example
How unusual is 190 cm?
Adult male height: roughly mean 175 cm, σ = 7.5 cm.
Distance from mean in σ-units (the z-score):
+2σ: about 97.7% of men are shorter. Same ruler works for test scores, lab values, anything bell-shaped — that's the point of z.
Out in the world
Polling margins of error
“52% ± 3%” is the bell curve at work: sample averages distribute normally around the truth (central limit theorem again), and ±3% marks the 95% window. Every election forecast leans on the math in this lesson.
Common confusion, cleared
“All data is normally distributed.”
Plenty isn't — incomes, city sizes, and earthquake magnitudes have long tails where bell rules fail badly. The theorem blesses sums and averages, not everything. Plot first.
“Two σ from the mean means something went wrong.”
About 5% of perfectly normal values live there. Rare isn't broken — a lesson medical patients (and their doctors) constantly relearn.
Recap
- Sums of many small influences pile into bells — that's the central limit theorem.
- 68 / 95 / 99.7 within ±1 / 2 / 3 σ: the universal ruler.
- Probability for continuous data is area under the curve.