Normal Distribution: Properties and Applications

The normal distribution is the bell-shaped, symmetric distribution that sits at the center of almost every introductory statistics course and underlies most of the inferential tests researchers actually use. It's defined entirely by two numbers, the mean and the standard deviation, and it has a small set of mathematical properties that make it the workhorse of statistical inference. Understanding what the normal distribution is, what it isn't, and when it matters is one of the highest-leverage things a researcher can learn early in their statistical training.

This guide covers what the normal distribution is, its defining properties, the 68-95-99.7 rule, the standard normal distribution and why it matters, the role the normal distribution plays in statistical inference, which variables actually follow it (and which don't), how to check whether your data is normal, and how to report normality in APA format. For the broader framework that descriptive statistics sit within, see the complete guide to descriptive statistics.

What Is the Normal Distribution?

The normal distribution is a continuous probability distribution that looks like a symmetric bell when plotted. Values cluster around the center, and the frequency of values drops off smoothly as you move away from the center in either direction. It's sometimes called the Gaussian distribution after Carl Friedrich Gauss, who developed many of its mathematical properties in the early 19th century.

The shape of the curve is determined by exactly two numbers: the mean, which sets where the center is, and the standard deviation, which sets how wide the bell is. A larger standard deviation produces a wider, flatter curve. A smaller one produces a narrower, taller curve. Two normal distributions with the same mean but different standard deviations have the same center but different spreads. Two with the same standard deviation but different means have the same shape but sit at different locations on the number line.

The total area under a normal distribution curve equals 1, because it's a probability distribution. The proportion of values that fall between any two points on the curve equals the area under the curve between those two points. This is what makes the normal distribution practically useful: once you know the mean and standard deviation, you can calculate the probability of getting any range of values.

The Defining Properties of the Normal Distribution

Four properties together define the normal distribution and distinguish it from other bell-shaped distributions like the t-distribution.

Property 1: Symmetric around the mean

The left half of the curve is a mirror image of the right half. The peak sits exactly at the mean. The proportion of values above the mean equals the proportion below it: 50% on each side. This is the property most people picture when they think of the bell curve.

Property 2: Mean equals median equals mode

For a perfectly normal distribution, the three measures of central tendency are identical. The mean sits at the center because the distribution is symmetric. The median sits at the center because exactly 50% of values fall on each side. The mode sits at the center because the peak of the curve is the most frequent value. This is one of the easiest ways to recognize that a distribution is normal: when mean, median, and mode all coincide.

Property 3: Defined entirely by mean and standard deviation

Two numbers describe a normal distribution completely. Once you know μ and σ, the entire shape is fixed. No additional information is needed. This is what makes the normal distribution mathematically convenient and why so much of inferential statistics is built around it.

Property 4: Asymptotic tails

The tails of the curve approach the horizontal axis but never quite touch it. In theory, the normal distribution extends infinitely in both directions. In practice, the probability of values more than three or four standard deviations from the mean is so small that researchers treat the tails as effectively bounded. A value six standard deviations from the mean has a probability of about one in a billion in a normal distribution.

The 68-95-99.7 Rule (Empirical Rule)

The single most useful fact about the normal distribution is that the proportion of values falling within a given number of standard deviations of the mean is fixed and predictable. This is known as the 68-95-99.7 rule, or the empirical rule.

• About 68% of values fall within one standard deviation of the mean (between μ − σ and μ + σ).
• About 95% of values fall within two standard deviations of the mean (between μ − 2 σ and μ + 2 σ).
• About 99.7% of values fall within three standard deviations of the mean (between μ − 3 σ and μ + 3 σ).

A concrete example. IQ scores are designed to follow an approximately normal distribution with a mean of 100 and a standard deviation of 15. The 68-95-99.7 rule tells you immediately that about 68% of people have IQ scores between 85 and 115, about 95% have IQ scores between 70 and 130, and about 99.7% have IQ scores between 55 and 145. You don't need a statistical table to know that. The empirical rule gives you the answer.

The rule also helps you spot unusual values. A score more than two standard deviations from the mean is rare (about 5% of values), and a score more than three standard deviations from the mean is exceptional (about 0.3% of values). This is the intuition behind much of what later becomes formal hypothesis testing. For a deeper explanation of the standard deviation itself, see Editor World's guide to standard deviation and variance.

The Standard Normal Distribution

The standard normal distribution is the special case where the mean is 0 and the standard deviation is 1. Any normal distribution can be transformed into the standard normal by subtracting the mean and dividing by the standard deviation. This transformation, called standardization, produces what's known as a z-score.

z-scores express how many standard deviations a given value sits from the mean. A z-score of 0 is exactly at the mean. A z-score of 1 is one standard deviation above the mean. A z-score of −1.5 is one and a half standard deviations below the mean. Because every normal distribution can be expressed in terms of z-scores, statisticians only need one table (the standard normal table) to calculate probabilities for any normal distribution.

The standard normal is also the foundation of most statistical software's distribution functions. When you ask SPSS, R, Stata, or Python for the probability of a value in a normal distribution, the program is converting your value to a z-score. It then looks up the probability in the standard normal. The standard normal is the universal currency of normal-distribution calculations.

Why the Normal Distribution Matters in Research

The normal distribution is the most consequential distribution in research statistics for two main reasons.

The Central Limit Theorem

The Central Limit Theorem is one of the most powerful results in statistics. It says the following. Take repeated random samples from any population (whatever the original distribution looks like) and calculate the mean of each sample. The distribution of those sample means will be approximately normal, even if the original population is not normal. The approximation gets better as the sample size increases.

This is why the normal distribution shows up everywhere in inferential statistics. The tests you run typically depend not on the original data being normal, but on the sampling distribution of the test statistic being approximately normal. The Central Limit Theorem usually makes that approximation safe at sample sizes above about 30, even when the underlying data is skewed.

Parametric test assumptions

Most parametric tests ( t-tests, ANOVA, Pearson correlation, linear regression) assume normality somewhere in their derivations. Sometimes the assumption applies to the underlying data. More often, it applies to the sampling distribution or to the residuals after the model is fit. Either way, the validity of the test depends on the assumption being approximately met.

Knowing whether your data or your model residuals are approximately normal is a routine part of choosing the right statistical test. When the assumption is badly violated, non-parametric alternatives (which don't assume normality) often work better.

What Actually Follows a Normal Distribution?

A common misunderstanding is that everything in nature follows a normal distribution. It doesn't. Many real-world variables follow other distributions, and some don't follow any neat distribution at all.

Variables that often approximate the normal distribution include heights of adults within a single sex and population, measurement errors in well-calibrated instruments, IQ scores (by design), reaction times in simple cognitive tasks (after appropriate transformation), and many biological measurements like blood pressure within a relatively homogeneous group.

Variables that do not follow the normal distribution include income, wealth, and most financial measures (right-skewed, with a long tail of high values), reaction times in raw form (right-skewed), counts of rare events (Poisson-shaped), and proportions (often beta-shaped or otherwise bounded). Fisher and Yao (2017) studied gender differences in financial risk tolerance using the Survey of Consumer Finances, a dataset where almost no financial variable follows a normal distribution. Household income, net worth, asset values, and debt are all positively skewed with heavy right tails, and researchers working with this kind of data routinely use medians, log transformations, or non-parametric methods rather than assuming normality.

The mean alone won't tell you whether your data is normally distributed. Two distributions with the same mean and standard deviation can look very different. Visualization is the right first step. For guidance on choosing the right visualization for the question you're asking, see Editor World's guide to box plots, scatter plots, and choosing the right visualization.

How to Check Whether Your Data Is Normal

There are three main approaches to assessing normality, and good practice uses at least two of them together.

Visual inspection

A histogram is the simplest check. If the histogram looks roughly bell-shaped, symmetric, and unimodal, the normal distribution is a reasonable working assumption. If the histogram is clearly skewed, has multiple peaks, or has heavy tails, the data is not normal.

A Q-Q plot (quantile-quantile plot) is more diagnostic. The plot shows the quantiles of your data against the quantiles of a theoretical normal distribution. If your data is normal, the points fall along a straight diagonal line. Departures from the line, especially at the tails, indicate departures from normality. Most statistical software produces Q-Q plots with a single command, and reviewers familiar with statistics interpret them quickly.

Summary statistics for shape

Skewness and kurtosis quantify departures from normality. For a normal distribution, skewness is 0 and excess kurtosis is 0. Many statistical packages report these values along with descriptive statistics. As a rough guideline, skewness values between −1 and 1 indicate approximate symmetry, while values beyond −2 or 2 indicate substantial skew.

Formal normality tests

Tests like the Shapiro-Wilk, Kolmogorov-Smirnov, and Anderson-Darling provide a formal test of the null hypothesis that the data comes from a normal distribution. In practice, these tests are sensitive to sample size: they almost always reject normality with large samples and often fail to detect non-normality in small ones. Most experienced researchers prefer visual diagnostics (histograms and Q-Q plots) supplemented by skewness and kurtosis, and use formal tests only when a specific decision requires them.

When Normality Matters and When It Doesn't

Beginning researchers often worry about normality more than they need to. The honest answer is that normality matters more in some situations than others.

Normality matters when sample sizes are small (below about 30) and the test depends on a normal sampling distribution. It matters when you're constructing prediction or tolerance intervals that depend on the exact shape of the distribution. It matters when departures are large enough to affect the p-value or the effect size estimate in a meaningful way.

Normality matters less when sample sizes are large. The Central Limit Theorem makes the sampling distribution of the mean approximately normal even when the underlying data is not. As a result, tests based on sample means (like the t-test) are reasonably robust at sample sizes above 30 or 50. Normality also matters less when you're using methods that don't assume it: non-parametric tests, robust statistics, or modern resampling methods like bootstrapping all sidestep the assumption entirely.

The right question isn't whether your data is normal in some absolute sense. It's whether the departures from normality are large enough to change the conclusions you'd reach. For most well-designed studies with reasonable sample sizes, the answer is no. For studies with small samples and badly skewed data, the answer can be yes, and a non-parametric or transformed approach is more defensible.

Reporting Normality in APA Format

When normality is relevant to your analysis (either because you assessed it before choosing a test, or because reviewers will ask whether the assumption was met), report it briefly in the analysis section.

For visual assessments, a sentence is enough. Example: "Inspection of histograms and Q-Q plots indicated that the dependent variable was approximately normally distributed within each group."

For formal tests, report the test statistic and p-value: "The Shapiro-Wilk test indicated no significant departure from normality (W = .98, p = .31)."

When normality is clearly violated and you've used a non-parametric or robust alternative, explain the choice: "Because the dependent variable showed substantial positive skew, a Mann-Whitney U test was used in place of the t-test."

Don't over-report. Most readers don't need a paragraph on every normality check you ran. A sentence or two confirming the assumption was reasonable, or explaining what you did when it wasn't, is usually enough.

Common Misunderstandings About Normality

• "Everything in nature is normally distributed." No. Many real-world variables are skewed, bounded, or otherwise non-normal. Income, wealth, reaction times, counts, and proportions are all non-normal in practice.
• "My data needs to be perfectly normal to use a t-test or ANOVA." No. These tests are robust to moderate departures from normality, especially with sample sizes above 30. The assumption applies more strictly to the sampling distribution than to the raw data.
• "A significant Shapiro-Wilk test means I can't use parametric methods." Not necessarily. Formal normality tests are sensitive to sample size and often reject normality for trivial departures in large samples. Visual diagnostics are usually more informative.
• "If the mean and median are close, the data is normal." Mean and median being close suggests symmetry, but symmetry isn't the same as normality. A symmetric distribution can have heavy or light tails that distinguish it from the normal.
• "Standardizing data makes it normal." Standardization (converting to z-scores) changes the mean to 0 and the standard deviation to 1 but doesn't change the shape of the distribution. Skewed data stays skewed after standardization.
• "The normal distribution is always the right model." Many other distributions (log-normal, beta, Poisson, exponential, gamma, Weibull) are appropriate for specific kinds of data. The normal is common but not universal.

When Professional Editing Helps

Statistics writing is one of the easiest places for small errors to undermine reviewer confidence. Mismatched notation, italics on the wrong elements, normality claims that aren't supported by the diagnostics, and misuses of "significant" all show up in nearly every first draft. Editor World's academic editing services include review of statistical notation, APA compliance, and the substantive accuracy of methodological claims. The same standard is applied across dissertation editing, journal article editing, and essay editing. 100% human editing, no AI at any stage. You choose your own editor from verified profiles, and a free sample edit is available before you commit. Browse available editors by subject expertise to find someone whose background matches your field.

Frequently Asked Questions About the Normal Distribution

Content reviewed by Editor World editorial staff. Editor World, founded in 2010 by Patti Fisher, PhD, graduate of The Ohio State University, provides professional editing and proofreading services for academic researchers, doctoral candidates, faculty, business professionals, and authors worldwide. BBB A+ accredited since 2010 with 5.0/5 Google Reviews and 5.0/5 Facebook Reviews. More than 100 million words edited for over 8,000 clients in 65+ countries. Stevie Award winner: Gold 2019, Bronze 2018 and 2025. Native English editors from the United States, the United Kingdom, and Canada with subject-matter expertise across the social sciences, the natural and physical sciences, medicine, engineering, computer science, and the humanities. 100% human editing, no AI at any stage. Less than 5% of applicants are accepted to the editor panel. Recommended by the Boston University Economics Department, University of San Diego, University of Michigan, UCLA, University of Missouri, and more.