Interpreting Departures in Normal Quantile Plots

What a Normal Quantile Plot Can Show

A normal quantile plot compares the ordered residuals from a regression fit with the quantiles expected from a normal distribution. Its value is in the shape of the departures from a reference line: a sustained bend or a tail that departs more than the other may indicate a particular way the residual distribution differs from normality. A single point away from the line, by itself, is weaker evidence.

For a consistent description, this tutorial uses a plot with theoretical normal quantiles on the horizontal axis and ordered residuals on the vertical axis. The reference line represents the pattern expected if the residuals were approximately normal, allowing for their location and scale. Some software reverses the axes, so always check which quantity is on which axis before describing the direction of a bend.

In a regression model, residuals are observed differences between responses and fitted values. A normal quantile plot is used to assess whether their distribution is reasonably consistent with the model’s normal-error assumption. It does not show the unobserved errors directly, and a plot cannot establish that the assumption is true.

Read the Reference Line Before the Bend

The reference line need not be the line \(y=x\). If the plot uses residuals in their original units and theoretical standard-normal quantiles, a normally distributed residual variable with mean \(\mu\) and standard deviation \(\sigma\) would follow approximately \(y=\mu+\sigma x\). The intercept and slope account for location and scale. Consequently, a straight pattern that is offset from zero or has a slope other than one is not, by itself, evidence of nonnormality.

The diagnostic question is whether the points follow an approximately straight pattern around the reference line. Small irregular departures are expected in finite samples. A sustained shape across several neighboring points is more informative than a single point, though its importance also depends on the sample size and the purpose of the analysis.

Symmetric Tail Departures

Compare the two ends of the plot to distinguish a symmetric tail pattern from an asymmetric one. With theoretical quantiles on the horizontal axis and residual quantiles on the vertical axis, heavier-than-normal tails tend to make the lower-tail points fall below the line and the upper-tail points rise above it. The observed extremes are farther from the central part of the distribution than the normal reference suggests. This often creates an outward-bending or S-like pattern.

Lighter-than-normal tails tend in the opposite direction: lower-tail points lie above the line and upper-tail points lie below it. The ends are comparatively close to the center. These are tendencies, not a rule that every finite-sample plot must follow exactly. The central points, tail length, and overall coherence of the bend all matter.

Asymmetric Departures and Skewness

When one tail departs more strongly than the other, the pattern is asymmetric. For example, a relatively short lower tail together with a longer, more extreme upper tail is consistent with right-skewness. Conversely, a more extreme lower tail and a comparatively short upper tail is consistent with left-skewness. The key comparison is the relative extent of the two tails, not simply whether points are above or below the line at one location.

The shape can be affected by the placement and slope of the reference line, so avoid diagnosing skewness from a memorized curve direction alone. First identify the axes and line; then compare how far the lower and upper extremes extend from the central pattern. A broad, coherent imbalance across several points is more persuasive than an unusual value at just one end.

A Short Quantile-Based Justification

Why does comparing the tails with the line give information about tail weight? A quantile describes the value below which a stated proportion of a distribution falls. For example, under the standard normal distribution, the quantiles at probabilities \(0.025\) and \(0.975\) are approximately \(-1.96\) and \(1.96\). In a normal quantile plot, sample residual quantiles at those approximate positions are compared with the normal reference values, adjusted for the line’s location and scale.

If the adjusted lower sample quantile is more negative than its normal counterpart and the adjusted upper quantile is more positive, both tails extend farther than the normal reference. In standardized units this comparison is:

where the \(q\)'s are the observed residual quantiles after accounting for the reference line. These inequalities describe more extreme tail quantiles at those probability positions, which supports the interpretation of heavier tails. If only the upper quantile is substantially more extreme, while the lower one is not, the evidence is asymmetric instead. This comparison is a justification for the visual reading, not a test with a specified significance level.

Patterns That Need Caution

A departure from the line is diagnostic evidence, not an explanation. The plot can suggest a distributional shape, but it does not show whether the cause is a few unusual observations, a feature of the data, or another issue with the fitted model. It also does not prove that the regression model’s other assumptions hold. Interpret the normal quantile plot alongside the analysis context and other appropriate diagnostics.

• One extreme point: A single point far from the line may indicate an unusual residual, but it does not by itself establish a systematic tail pattern.
• A straight line with a different slope: A straight pattern may reflect a different scale from the standard normal. Judge departures from the reference line, not from a fixed slope of one.
• Small waviness: Ordered sample values will not fall perfectly on a line. Do not assign a strong distributional interpretation to minor irregularities.
• One tail only: Describe which tail departs and how. Do not call a one-sided departure symmetric heavy tails.

A common mistake is to say, “The upper points are above the line, so the residuals have heavy tails.” That reasoning considers only one end. Heavy-tail evidence is stronger when the lower tail also departs outward in the corresponding direction. If the upper tail alone is unusually extended, the pattern may instead be asymmetric.