Normality in statistics means that a dataset follows a normal distribution—a symmetric, bell-shaped curve where most values cluster around the mean and fewer appear at the extremes. You check normality to decide whether parametric tests (like t-tests and ANOVA) are valid, or to justify using the empirical rule for probabilities. This article explains what normality is, why it matters, how to test for it, and what to do when your data are not normal.
Formula
The normal distribution is defined by its probability density function:
f(x) = (1 / (σ √(2π))) · e−((x − μ)²)/(2σ²)
- μ (mu) – the mean, the center of the distribution (unit of the data).
- σ (sigma) – the standard deviation, the spread (same unit as the data).
- x – any value of the variable.
- π ≈ 3.14159 and e ≈ 2.71828.
The z-score standardizes any value: z = (x − μ) / σ. It tells you how many standard deviations x is from the mean. The calculator above computes the z-score, the cumulative probability (percent of data below x), and the standard error of the mean (σ / √n) for a given sample.
Worked Example
Suppose a test has a mean of 100 and a standard deviation of 15 (like many IQ tests). You want the probability that a randomly chosen score is below 120, and the standard error for a sample of 30 people.
- Z-score: z = (120 − 100) / 15 = 20 / 15 = 1.3333
- Percent below 120: using the cumulative normal function, Φ(1.3333) ≈ 0.9088, so 90.88%.
- Standard error: SE = 15 / √30 = 15 / 5.4772 = 2.7386
The calculator returns z = 1.3333, percent below = 90.88%, and SE = 2.7386, matching these steps.
Assumptions
- The data are continuous and measured on an interval or ratio scale.
- The distribution is exactly symmetric and bell-shaped (no skew or excess kurtosis).
- The sample is independent and randomly drawn from the population.
- For the standard error formula, the sample is a small fraction of the population (or sampling is with replacement).
Why Normality Matters
Many classical statistical procedures assume normality. The t-test, ANOVA, linear regression, and confidence intervals based on the t-distribution all rely on the residuals (or the sampling distribution of the mean) being approximately normal. When the assumption holds, these tests have known properties: controlled Type I error rates and maximum power. When it fails, p-values can be misleading—too small or too large—leading to wrong conclusions.
However, thanks to the Central Limit Theorem, the sampling distribution of the mean approaches normality as sample size grows, even if the underlying data are not normal. For n ≥ 30 (sometimes n ≥ 50 for heavily skewed data), t-tests and ANOVAs are usually robust. So normality of the raw data matters most for small samples, for prediction intervals, and for methods like maximum likelihood estimation that assume a normal likelihood.
How to Check Normality
There are three complementary ways: graphical, numerical, and formal tests.
- Histogram or density plot – look for a single peak, symmetry, and no extreme outliers.
- Q-Q plot – plot your sorted data against theoretical normal quantiles. Points close to a straight diagonal line suggest normality.
- Skewness and kurtosis – skewness near 0 and kurtosis near 3 (or excess kurtosis near 0) indicate symmetry and tail weight similar to normal.
- Shapiro-Wilk test – the most powerful formal test for normality, works well for n < 5000.
- Kolmogorov-Smirnov test – compares your sample to a fully specified normal distribution (requires estimating parameters, so use Lilliefors correction).
- Anderson-Darling test – sensitive to deviations in the tails.
In practice, graphical checks are often enough. Formal tests can be overly sensitive with large samples—a tiny deviation from normality may be statistically significant but practically irrelevant.
Common Mistakes and Misconceptions
- Confusing normality of data with normality of residuals – In regression, you only need the residuals to be normal, not the raw outcome variable.
- Assuming normality means “no outliers” – The normal distribution has no true outliers; extreme values are just unlikely. But real data often have genuine outliers that break normality.
- Using a test on a tiny sample – With n < 10, normality tests have very low power; they can miss severe non-normality. Always plot the data too.
- Ignoring the Central Limit Theorem – For large samples, you can often proceed with parametric tests even if the data are skewed.
- Transforming data without checking – Log or square-root transformations can help, but they change the interpretation of results.
What to Do If Data Are Not Normal
If normality fails, you have several options:
- Use nonparametric tests – Mann-Whitney U instead of independent t-test, Wilcoxon signed-rank instead of paired t-test, Kruskal-Wallis instead of one-way ANOVA. These tests do not assume normality, only similar shapes.
- Transform the data – Log, square root, or Box-Cox transformations can make skewed data more symmetric.
- Use robust methods – Trimmed means, bootstrapping, or rank-based procedures that are less sensitive to outliers.
- Model with a different distribution – For count data, use Poisson or negative binomial regression; for skewed positive data, use gamma or log-normal models.
- Increase sample size – With enough data, the Central Limit Theorem makes the sampling distribution normal, so parametric tests become valid.
Frequently Asked Questions
Is normality the same as a bell curve?
Yes, the normal distribution is often called the bell curve because its density function looks like a bell. But not every bell-shaped curve is normal; other distributions (like the t-distribution with few degrees of freedom) also look bell-shaped but have heavier tails.
Can I use t-test if my data are skewed but n is large?
Generally yes. The Central Limit Theorem ensures the sample mean is approximately normal for large n (often n ≥ 30, but more for severe skew). The t-test will have accurate Type I error rates, though it may have slightly lower power than a nonparametric alternative.
What is the difference between normal distribution and standard normal distribution?
The standard normal distribution has a mean of 0 and a standard deviation of 1. Any normal distribution can be converted to standard normal by subtracting the mean and dividing by the standard deviation—that is exactly the z-score transformation.
How large a sample do I need to assume normality?
There is no fixed number, but a common rule of thumb is n ≥ 30 for moderate skew, and n ≥ 50 or more for heavily skewed data. For very small samples (n < 10), normality is a strong assumption that should be verified with plots, not tests.
Does normality apply to the dependent variable or the residuals?
In regression and ANOVA, the assumption is about the residuals (errors), not the raw dependent variable. For a simple t-test comparing two groups, it is the distribution of the outcome within each group that matters, but the test is robust for large samples.
What if my data have outliers—can I still claim normality?
Outliers are the most common cause of non-normality. If you have a few extreme values, you can investigate them (are they errors?) and consider robust methods. Removing outliers just to achieve normality is not acceptable unless they are genuine measurement errors.
Related
Formula used
in m | Sample mean | - | 100 | 0 in sd | Sample standard deviation | - | 15 | 0.0001 in x | Value of interest | - | 120 | 0 in n | Sample size | - | 30 | 2 out z | Z-score | - | (x - m) / sd | 4 out p | Percent below x (normal model) | % | 100 * 0.5 * (1 + erf(z / sqrt(2))) | 2 out se | Standard error of the mean | - | sd / sqrt(n) | 4
Each line is one input (in, sel, list) or one result (out: label, unit, formula, decimals). How the calculators work
Method: the standard formula for this calculation with the constants and conversion factors written on the page, rounded as each result states. See method and constants, the disclaimer or report an error.
More in this section
- Normality to g/L Converter & Formula29.220 g/L
- Normality to meq/mL Conversion Calculator1.0000 meq/mL
- Normality Test Q-Q Plot: How to Read It
- Normality Test Results: How to Analyze Them
- Normality Slay the Spire?
- Normality Test for Paired t Test: How to Check