The Normal Distribution

What it is

The normal distribution — also called the Gaussian distribution, or just the bell curve — is the most important shape in statistics. Values cluster near a central point and thin out symmetrically as you move away in either direction, so the curve is tall in the middle and tapers smoothly toward zero in both tails. It is fully described by two parameters: the mean μ, which sets where the peak sits, and the standard deviation σ, which sets how wide the curve spreads.

Because it's perfectly symmetric, a normal distribution has its mean, median, and mode all equal at the center. Its skewness is 0 (no lean to either side) and its excess kurtosis is 0 (the reference against which other shapes are called heavy- or light-tailed). Slide μ and the whole curve shifts left or right; change σ and it gets narrower and taller, or flatter and wider, but the bell shape never changes.

The formula

The probability density function — the height of the curve at any value x — is

f(x) = ( 1 / (σ·√(2π)) ) · e^( −(x − μ)² / (2σ²) )

The two parameters carry all the meaning:

The (x − μ)² term is what makes the curve symmetric: a value above the mean and one the same distance below get identical heights. The σ² in the denominator is what makes a larger spread flatten the curve.

How to read it

The single most useful fact about the normal distribution is the 68–95–99.7 rule, also called the empirical rule. It says where the data sit relative to the mean, measured in standard deviations:

Read it as a quick test of how typical a value is: anything beyond 2σ from the mean is in the outer 5%, and anything beyond 3σ is in the outer 0.3% — rare enough to be worth a second look. To express a specific value's distance from the mean in standard deviations, you compute its z-score.

Why it matters

Many of the most-used methods lean on normality. The t-test, ANOVA, and the inference behind linear regression (its confidence intervals and p-values) all assume the data — or, more often, the model's residuals — are approximately normal. The reason the assumption is so often reasonable is the Central Limit Theorem: averages of many independent values trend toward a normal distribution even when the raw data don't. That's why sample means, totals, and regression coefficients are so often well-behaved.

It also means "is my data roughly normal?" is a question worth answering before you trust those tests. Two checks do the job: a histogram with a fitted normal curve overlaid shows departures from the bell shape directly, and a Q-Q plot — which plots your sorted data against the values a normal distribution would predict — turns normality into a straight-line test. Points that hug the diagonal mean normal; systematic curves away from it mean skew or heavy tails.

The normal distribution is the reference shape the rest of statistics is built against. Learn to read it in standard deviations — that's what the z-score formalizes — and you can judge any normally distributed value at a glance.