What is the standard normal distribution?

The normal distribution with mean zero and standard deviation one.

Does normal mean no outliers?

No. Extreme values can occur, and their frequency depends on the model and sample size.

Is a normal model appropriate for counts?

It may sometimes approximate certain count calculations, but counts are discrete and require attention to bounds and approximation quality.

Are mean and standard deviation enough to prove normality?

No. Many distributions share those two summaries.

AI Glossary / Statistics and probability / Beginner

Normal distribution

Definition

The normal distribution, also called the Gaussian distribution, is a continuous probability distribution with a symmetric bell-shaped density. Its mean mu determines the center, and its positive standard deviation sigma determines the spread. It is widely used to model measurement variation, errors, and sampling distributions under appropriate conditions. The normal distribution is a specific mathematical family, not a general label for data that seem ordinary or approximately balanced.

What it means

Values near the center have greater density than values far away, and equal distances on opposite sides of the mean have the same density. The mean, median, and mode coincide. The model extends over all real numbers, so its suitability must be considered when the real quantity has strict bounds or strong asymmetry.

A density's height is not the probability of one exact value. For a continuous normal variable, the probability of any single exact value is zero. Probabilities correspond to areas across intervals, which can be calculated using the cumulative distribution function.

How it works

Specify the mean and standard deviation, either as known model quantities or estimates with appropriate uncertainty. Convert a value x into a standardized score z = (x − mu)/sigma. This expresses its distance from the mean in standard-deviation units.

Use the standard normal cumulative distribution to obtain lower-tail probabilities, upper-tail probabilities, or differences between cumulative probabilities for intervals. Approximately 68.27% of a normal distribution lies within one standard deviation of its mean, 95.45% within two, and 99.73% within three. About 95% lies within 1.96 standard deviations.

Check whether the normal approximation suits the intended calculation. A central limit theorem can justify approximate normality for certain standardized averages under suitable conditions; it does not make the original observations normal. Dependence, heavy tails, small samples, and extreme-tail questions can affect whether a particular approximation is reliable.

Three practical examples

An engineering team uses a normal model for small measurement errors after checking its fit and stability.
A business estimates uncertainty in a sample average when the sampling design and approximation conditions support the calculation.
An AI researcher generates Gaussian perturbations in a simulation, explicitly choosing their mean and standard deviation.

Think of it this way

Think of a smooth hill viewed from the side: its center locates the highest point, while its width describes how spread out the distribution is. The probability of an interval is the area beneath that stretch of the hill, not the height at a single point.

Worked example

Consider an original teaching model for a measurement X with mean 100 units and standard deviation 8 units. Assume X is exactly normally distributed. A value of 116 has z = (116 − 100)/8 = 2, so the probability of exceeding 116 is approximately 0.02275, or 2.275%.

The interval from 84 to 116 corresponds to z values −2 and +2. Its probability is approximately 0.95450, or 95.45%. A central 95% interval for individual measurements is approximately 100 ± 1.96 × 8, giving [84.32, 115.68].

Now take the average of 16 independent measurements from this same normal population. Its mean remains 100, while its standard deviation is 8/sqrt(16) = 2. An average of 104 is two standard errors above 100, whereas an individual measurement of 104 is only half a standard deviation above 100. The distinction shows why the spread of individual outcomes must not be confused with the spread of sample averages. All probabilities here follow from the stated model; they are not empirical findings about a real measurement process.

Illustrative case study

Fictional case study: Cedar Instruments summarizes repeated calibration measurements with a normal model. Its dashboard initially uses the standard error of the mean as though it described the variability of a single future reading, producing an unrealistically narrow displayed range.

An analyst separates the distribution of individual readings from uncertainty about their average. Cedar updates the labels and evaluates whether the normal model adequately represents its observed tails. The fictional case illustrates that even a familiar distribution can produce misleading communication when the quantity being described is unclear.

Common misconceptions and mistakes

“All large datasets become normal.” A larger dataset can reveal a non-normal distribution more clearly. A theorem about averages is not a theorem about raw values.

“Symmetric means normal.” Many symmetric distributions have different shapes and tail behavior.

“Two standard deviations always covers exactly 95%.” For an exact normal distribution the figure is about 95.45%; other distributions differ.

“Standardization makes data normal.” Subtracting a mean and dividing by a standard deviation changes location and scale, not the underlying shape.

Business use

Normal models support many practical calculations, but a business should first identify what is being modeled: individual demand, measurement error, a residual, or an estimator. These objects can have different distributions even within the same analysis.

Use graphical checks and substantive knowledge alongside numerical diagnostics. Be especially careful with tail probabilities used for capacity planning, reliability, or exception thresholds, because a modest departure in the center can coexist with a large error in the tails. Where the approximation is inadequate, consider a different distribution or a method that better matches the data and decision.

Check your understanding

A normal variable has mean 40 and standard deviation 5. What standardized score corresponds to 30?

Answer: (30 − 40)/5 = −2. The negative sign indicates a value below the mean, and the magnitude indicates a distance of two standard deviations.

Related terms

Cumulative distribution
Z-score
Standard deviation
Standard error
Central limit theorem
Gaussian distribution
standard normal
probability density

Learn next

Practice translating intervals into standardized scores and probabilities. Then compare histograms and quantile plots for normal and non-normal data. Study the sampling distribution of an average separately from the distribution of individual observations, including the assumptions behind normal approximations.

Sampling distribution

Sources

NIST/SEMATECH e-Handbook, normal distribution, density and parameters: https://www.itl.nist.gov/div898/handbook/eda/section3/eda3661.htm

Frank Harrell, Glossary of Statistical Terms, normal distribution and estimator-focused interpretation: https://hbiostat.org/glossary/

The measurement example and Cedar Instruments case are original illustrations. Their numerical probabilities assume the explicitly specified normal model.

Previous: Normal Approximation To The Binomial Distribution Next: Normal distributions transform