The Normal Distribution: The Bell Curve Explained
The normal distribution, or bell curve, is symmetric around its mean. Learn the 68-95-99.7 rule, z-scores, why it appears so often, and where it fails.
The normal distribution, also called the Gaussian distribution or bell curve, is a symmetric, single-peaked probability distribution defined by just two numbers: its mean, which sets the centre, and its standard deviation, which sets the spread. Values close to the mean are common, and values far from it are rare, falling off smoothly on both sides. It is the most widely used distribution in statistics, because many natural measurements cluster this way and because the sum of many small random effects tends to look normal.
What is the shape of a normal distribution?
The curve is perfectly symmetric, so the mean, median, and mode all sit at the peak. Changing the mean slides the curve left or right; changing the standard deviation makes it wider or narrower. The total area under the curve is 1, which represents 100 percent probability. The height of the curve at any point x is given by the density formula:
f(x) = 1 / (σ × sqrt(2π)) × exp( −(x − μ)² / (2σ²) )
Here μ (mu) is the mean and σ (sigma) is the standard deviation. The tails stretch out forever but shrink so quickly that values far from the mean are vanishingly unlikely.
The 68-95-99.7 rule
A handy rule of thumb says how much of the data falls within a given number of standard deviations of the mean:
| Range | Share of values |
|---|---|
| Within 1 standard deviation | about 68.3 percent |
| Within 2 standard deviations | about 95.4 percent |
| Within 3 standard deviations | about 99.7 percent |
More precisely, exactly 95 percent of values lie within 1.96 standard deviations, the number behind the familiar 95 percent confidence interval. As an example, many intelligence tests are scaled to a mean of 100 and a standard deviation of 15, so about 68 percent of people score between 85 and 115 and about 95 percent score between 70 and 130.
What is a z-score?
A z-score says how many standard deviations a value is from the mean: z equals the value minus the mean, divided by the standard deviation. A z-score of 0 is exactly average, +1 is one standard deviation above, and −2 is two below. Converting to z-scores puts different scales on the same footing, so you can compare a height with an exam mark, and it lets software look up the probability of any range. A score of 130 on the scale above has a z-score of 2, which is higher than about 97.7 percent of people.
Why does the bell curve appear so often?
The reason is the central limit theorem: when many small, independent effects add up, the total tends toward a normal distribution, whatever the shape of the individual effects. Human height is shaped by many genes and many environmental factors, and measurement errors come from many tiny disturbances, so both come out close to normal. The same theorem is why averages of samples are approximately normal even when the raw data are not.
The curve has a long history. Abraham de Moivre used it in 1733 to approximate the results of many coin flips. Carl Friedrich Gauss used it in 1809 to describe errors in astronomical measurements, which is why it carries his name. Francis Galton's bean machine, in which balls bounce left or right through rows of pegs, shows the shape appearing in a physical device.
Where the bell curve fails
Not everything is normal, and assuming it is can be costly. Incomes, city populations, and word frequencies are heavily skewed, with long tails of very large values. Earthquake energies and wealth follow power laws. Financial returns have fat tails: under a normal model a five-sigma daily move should occur about once in many thousands of years of trading days, yet real markets have produced several within a few decades. A few extreme values can also distort the mean and standard deviation, so analysts check normality with plots and tests, or switch to methods that do not depend on it.
Many classic procedures, including t-tests and confidence intervals, assume approximately normal data or large samples, which is why the idea sits behind most hypothesis testing. Quality-control programmes that speak of six sigma use the same curve: the label refers to a process so consistent that defects are rare, around 3.4 per million opportunities under the usual convention.
Tags
bell curve data analysis distributions probability statistics