The normal distribution is the bell-shaped curve that turns up whenever many small, independent influences add together. Heights, measurement errors and test scores often follow it closely. It is also the shape that sample means tend towards, which is why so many tests rely on it.
Two numbers describe it completely
A normal distribution is fixed by its mean μ (where the peak sits) and its standard deviation σ (how wide the bell is). Every normal curve has the same shape. Change μ and the curve slides along. Change σ and it stretches or squashes.
The 68-95-99.7 rule
For any normal distribution:
- 68.3% of values lie within 1 σ of the mean
- 95.4% lie within 2 σ
- 99.7% lie within 3 σ
Suppose adult heights in a population are normal with mean 170 cm and SD 10 cm. Then about two thirds of adults are between 160 and 180 cm, and only about 3 in 1,000 are outside 140 to 200 cm.
Answering questions with z-scores
To find a probability for any normal distribution, convert to a z-score and use the standard normal (mean 0, SD 1):
z = (x − μ) ÷ σ
What share of adults are taller than 185 cm? z = (185 − 170) ÷ 10 = 1.5. The z table gives 0.93319 to the left, so 1 − 0.93319 = 0.0668. About 6.7%.
Work this out in the normal distribution calculator
How tall do you need to be to be in the tallest 10%? This goes the other way: from an area to a value. The z with 90% below it is 1.2816, so the height is 170 + 1.2816 × 10 = 182.8 cm. The inverse normal calculator does this directly.
Why it matters even when your data aren’t normal
The central limit theorem says that the mean of a reasonably large sample is close to normally distributed, whatever shape the original data have. Waiting times are skewed, but the average of 50 waiting times follows a near-normal curve. This is why t-tests and confidence intervals work well for large samples even with skewed data.
“Reasonably large” depends on how skewed the data are. For mild skew, 30 is plenty. For very skewed data, such as incomes, you may need several hundred.
Checking whether data are normal
Look first, then test. A histogram should look roughly like a symmetric bell. A Q-Q plot should show points close to a straight line. The Shapiro-Wilk test gives a formal p-value, but be careful with it: with large samples it flags tiny departures that don’t matter, and with small samples it misses large ones that do.
Check yourself
IQ scores have mean 100 and SD 15. What share of people score above 130?
z = (130 − 100) ÷ 15 = 2. The area above z = 2 is 0.0228, so about 2.3%.
A normal distribution has mean 50. 95% of values lie between 30 and 70. Roughly what is σ?
95% is about 2 σ either side, so 2σ ≈ 20 and σ ≈ 10. More exactly, 1.96σ = 20 gives σ = 10.2.