AI Glossary / Statistics and probability / Foundation
Standard normal distribution
Definition
The standard normal distribution is the normal, or Gaussian, probability distribution with mean zero and variance one. Its standard deviation is therefore also one, and it is commonly written N(0,1). It provides a shared reference for probabilities, quantiles and many statistical approximations. A variable with mean zero and variance one is not necessarily standard normal: the normal shape is an additional, essential part of the definition.
What it means
Normal distributions differ in their center and spread but have the same basic bell-shaped form. Subtracting the mean and dividing by the positive standard deviation converts a genuinely normal variable to the standard normal scale. Values on that scale describe distance from the mean in standard-deviation units. This lets many probability calculations use one reference distribution. It does not mean that applying the same arithmetic to arbitrary data makes those data normal.
How it works
If X is normal with mean mu and standard deviation sigma greater than zero, define Z = (X minus mu)/sigma. Then Z follows N(0,1). The standard normal density at z is exp(-z squared/2) divided by the square root of 2 pi. Probabilities come from areas under that density, commonly calculated using the cumulative distribution function Phi. For example, P(a < Z < b) equals Phi(b) minus Phi(a). Use the inverse cumulative function to find a cutoff for a specified probability, and check which tail a table or software function returns.
Three practical examples
Example 1: A process analyst calculates a tail probability under an explicitly assumed normal model for a measurement. Example 2: A researcher uses a justified normal approximation to an estimator's sampling distribution. Example 3: An AI simulation draws standard normal noise and transforms it into a required Gaussian input. These uses concern different objects: raw measurements, estimates and simulated values. A normal approximation for an average does not automatically imply that individual observations have a normal distribution.
Think of it this way
Think of converting distances from different maps to a common ruler measured in local spread units. A value one unit above the center means one standard deviation above the mean, regardless of the original units. For normal distributions, the same ruler also gives the same probability areas. For a differently shaped distribution, matching center and spread is not enough to reuse those normal probability areas.
Worked example
Suppose a measurement X is modeled as normal with mean 100 and standard deviation ten. To find the probability that it lies between 90 and 115, standardize the endpoints: (90 minus 100)/10 = -1 and (115 minus 100)/10 = 1.5. Standard normal cumulative probabilities are approximately Phi(-1) = 0.158655 and Phi(1.5) = 0.933193. Their difference is 0.774538, or about 77.45 percent. The probability of exceeding 115 is 1 minus 0.933193 = 0.066807, about 6.68 percent. These results are conditional on the stated normal model and parameters. They are not guaranteed empirical proportions in a finite sample, and they should not be applied to a strongly skewed population simply because its mean and standard deviation happen to be 100 and ten.
Illustrative case study
In this fictional case study, Cedar Measurement uses a normal model to summarize a stable instrument's readings. Analysts compare the model with observed distributional patterns and investigate unusually heavy tails before estimating rare-event probabilities. Their report separates the arithmetic based on the normal curve from the evidence supporting that curve as an approximation. When an operating condition changes, they reassess the model rather than assuming that the same standardized cutoff preserves its earlier practical meaning.
Common misconceptions and mistakes
Common mistakes include confusing density with probability at a point, using the wrong tail of a normal table, and treating zero mean and unit variance as proof of normality. For a continuous normal variable, the probability of one exact value is zero even though the density there can be positive. Another error is automatically using normal critical values when estimated uncertainty and a small-sample model call for a different distribution. The central limit theorem has conditions and concerns sampling distributions; it does not transform every raw dataset into a bell curve.
Business use
Businesses use the standard normal reference to communicate model-based probabilities and support statistical methods, but the quality of the underlying assumptions matters more than numerical precision. Check whether the model describes the quantity of interest and whether extreme tails are supported by evidence. Values far from the center may be particularly sensitive to departures from normality. When reporting thresholds, name the distributional assumption and distinguish a model probability from an observed frequency or a decision requirement.
People also ask
Is the standard normal mean always zero?
Yes, by definition.
Is its variance one?
Yes.
Does a z-score of one mean a probability of one?
No; it is a position on the scale.
Does standardizing a skewed variable make it normal?
No.
What is Phi(0)?
One-half, by symmetry.
Is approximately 1.96 the upper 97.5th percentile?
Yes; it marks a commonly used normal reference cutoff.
Check your understanding
What standardized value corresponds to 115 in the example?
Answer: 1.5.
What is the probability between 90 and 115 under the stated model?
Answer: Approximately 0.774538.
Why subtract the two cumulative probabilities?
Answer: The subtraction leaves the area between the endpoints.
Would the same calculation necessarily work for a skewed variable with matching mean and spread?
Answer: No.
Related terms
- Z-score
- Cumulative distribution function
- Quantile
- Related terms include Gaussian distribution
- normal approximation
- standardisation and Student's t distribution
Learn next
Learn next how cumulative and survival probabilities correspond to different tails, and how quantiles reverse that calculation. Compare a normal variable with a nonnormal variable having the same mean and variance. Then study when a sampling distribution is approximately normal and when parameter estimation changes the appropriate reference procedure.
More in Statistics and probability
- Standard error - The standard error is the standard deviation of an estimator's sampling distribution.
- Standard error of differences between means - The standard error of differences between means is the standard deviation of the sampling distribution of a difference between estimated means.
- Standard lottery - A standard lottery is a reference gamble used in expected-utility assessment, typically offering a defined best outcome with probability p and a defined...
- Standard normal random vector - A standard normal random vector is a vector whose components are independent standard normal random variables.
- Standard scores - Standard scores express observations on a reference scale with a specified mean and standard deviation.
- Standardization - Standardization, in the statistical preprocessing sense used here, transforms a variable by subtracting a chosen mean and dividing by a chosen positive...
Sources
Sources: NIST/SEMATECH Engineering Statistics Handbook, Normal Distribution, https://itl.nist.gov/div898/handbook/eda/section3/eda3661.htm; Cumulative Distribution Function of the Standard Normal Distribution, https://www.itl.nist.gov/div898/handbook/eda/section3/eda3671.htm. These primary institutional references define the distribution and probability calculations. The interval example is original and independently calculated; the measurement case is fictional.
Previous: Standard lottery Next: Standard normal random vector