A Q-Q plot (quantile-quantile plot) is a scatterplot that compares your sample data against a theoretical normal distribution; if the points follow the diagonal reference line, your data are approximately normal. You read it by looking at how closely the plotted points hug that straight line, with systematic curves or S-shapes indicating skewness or heavy tails. It is one of the most reliable visual tools for checking the normality assumption before running t-tests, ANOVA, or linear regression.
Formula
The Q-Q plot itself does not have a formula—it is a graphical method. But the normality index computed here is a quantitative companion based on the Jarque–Bera test statistic:
JB = n × (S² / 6 + K² / 24)
where:
- n = sample size (number of observations)
- S = sample skewness (dimensionless, 0 for a symmetric distribution)
- K = sample excess kurtosis (dimensionless, 0 for a normal distribution)
The p-value shown is an approximation for small samples; the exact chi-square distribution with 2 degrees of freedom is used in statistical software. A p-value below 0.05 suggests the data deviate significantly from normality.
Worked Example
Suppose you have a sample of n = 50 observations, and you compute a skewness of 0.5 and an excess kurtosis of 0.8. Plugging into the formula:
JB = 50 × (0.5² / 6 + 0.8² / 24) = 50 × (0.25/6 + 0.64/24) = 50 × (0.041667 + 0.026667) = 50 × 0.068333 = 3.4167
The p-value approximation is exp(−3.4167/2) = exp(−1.7083) ≈ 0.181. That is above 0.05, so you would not reject normality—consistent with a Q-Q plot that shows only minor wiggles around the line.
Assumptions
- The Q-Q plot assumes the theoretical distribution (normal) is fully specified by its mean and standard deviation, which are estimated from the sample.
- The normality index assumes the sample is independent and identically distributed.
- The p-value approximation works best for n > 20; for very small samples, use exact tables.
How a Q-Q Plot Is Constructed
To build a Q-Q plot by hand, you sort your data from smallest to largest and assign each point a percentile rank. The common formula for the i-th smallest value in a sample of size n is (i − 0.5) / n, which gives the proportion of data below that point. You then convert that proportion to a z-score using the inverse normal cumulative distribution function (e.g., the z-score for 0.5 is 0, for 0.975 is 1.96). Plot the sorted data values on the y-axis and the theoretical z-scores on the x-axis. If your data come from a normal distribution, the points will fall roughly on the line y = μ + σ·z, which appears as a straight diagonal line.
Most software (R, Python, SPSS, Minitab) does this automatically, but understanding the construction helps you interpret what you see. The reference line is usually drawn through the first and third quartiles of both datasets, which makes it robust to outliers.
Reading the Plot: What the Patterns Mean
The key is to look at the shape of the points relative to the diagonal line:
- Points hug the line – data are approximately normal.
- Points curve upward at the right end (concave up) – right-skewed (positive skew): the upper tail is longer than the lower tail.
- Points curve downward at the left end (concave down) – left-skewed (negative skew): the lower tail is longer.
- Points form an S-shape (low points below line, high points above) – heavy tails (leptokurtic): more extreme values than normal.
- Points form an inverted S-shape (low points above line, high points below) – light tails (platykurtic): fewer extreme values than normal.
- One or two points far from the line at either end – outliers, which can distort the line.
Remember that small samples (n < 30) will show random scatter even when perfectly normal. The plot is most reliable with n of 50 or more.
Common Mistakes and Misconceptions
A frequent error is over-interpreting minor deviations. Real data are never perfectly normal; the question is whether the deviation is large enough to matter. A Q-Q plot with a slight wiggle but no systematic curve is usually fine for parametric tests, which are robust to mild non-normality when sample sizes are moderate.
Another mistake is using the wrong reference line. Some software draws the line through the mean and standard deviation, others through quartiles. The latter is more robust to outliers, but both are valid—just be consistent when comparing plots.
People also confuse the Q-Q plot with the P-P plot (probability-probability plot). The P-P plot compares cumulative probabilities directly and is more sensitive to deviations in the center of the distribution, while the Q-Q plot is better at showing deviations in the tails. For normality checking, the Q-Q plot is generally preferred.
Finally, a Q-Q plot does not tell you whether your data are exactly normal—it only shows how close they are. Use it alongside a histogram and a formal test (Shapiro–Wilk, Anderson–Darling) for a complete picture.
When to Use the Q-Q Plot
Use a Q-Q plot whenever you plan to apply a statistical method that assumes normality:
- Before a one-sample or two-sample t-test
- Before ANOVA (the residuals, not the raw data, should be checked)
- Before linear regression (check the residuals)
- When estimating confidence intervals based on the normal distribution
- When checking if a transformation (log, square root) is needed
For regression and ANOVA, always plot the residuals, not the original outcome variable. The normality assumption applies to the errors, not the marginal distribution of the response.
Alternatives and Complements
While the Q-Q plot is visual, you can confirm its message with numerical tests:
- Shapiro–Wilk test – powerful for small to moderate samples (n < 5000).
- Kolmogorov–Smirnov test – compares to a fully specified normal distribution; less powerful because parameters are estimated.
- Anderson–Darling test – gives more weight to the tails.
- Jarque–Bera test – based on skewness and kurtosis; the formula above.
None of these tests is definitive. With large samples, even trivial deviations become statistically significant, so always combine the test result with the visual Q-Q plot and your knowledge of the subject matter.
Frequently Asked Questions
What does a Q-Q plot tell you about normality?
A Q-Q plot shows whether your data’s quantiles match the quantiles of a normal distribution. If the points fall along the diagonal line, the data are consistent with normality; systematic deviations indicate skewness, heavy tails, or outliers.
How do I know if my Q-Q plot is good enough?
Look for a random scatter around the line with no clear curve. Minor wiggles are fine, especially with small samples. If the points stay within a band roughly parallel to the line, normality is a reasonable assumption.
What if my Q-Q plot shows an S-shape?
An S-shape (low points below, high points above) indicates heavy tails—more extreme values than a normal distribution. This often means the data have outliers or come from a distribution like Student’s t. Consider robust methods or a transformation.
Can I use a Q-Q plot for non-normal data?
Yes, you can plot your data against any theoretical distribution (e.g., exponential, uniform). The same logic applies: if the points follow the line, the data match that distribution. The normal Q-Q plot is just the most common.
Does sample size matter for Q-Q plots?
Yes. With small samples (n < 20), even perfectly normal data can show large random deviations. With large samples (n > 200), the plot becomes very sensitive, and minor deviations may appear that are not practically important.
Should I check normality on raw data or residuals?
For regression and ANOVA, check the residuals (observed minus predicted). The assumption is that the errors are normally distributed, not the outcome variable itself. For a simple one-sample test, check the raw data.
Related
Formula used
in n | Sample size | - | 50 | 3 in skew | Skewness (dimensionless) | - | 0 | -10 in kurt | Excess kurtosis (dimensionless) | - | 0 | -5 out z | Z-score for normality index | - | (skew^2 / 6 + kurt^2 / 24) * n | 2 out p | Approx. p-value (chi-square df=2) | - | exp(-z/2) | 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
- ug to ppb Calculator: Convert Micrograms per Liter1.00 ppb
- Grams to Moles Chemistry: How to Convert0.2000 mol/L
- Convert ug/l to ppb Calculator
- CPP Calculator Ontario: How CPP Is Calculated162.12 $/month
- Grams to Moles to Molecules: How to Convert0.2000 mol/L
- Organic Chemistry Reaction Calculator: How It Works