Q-Q Plots: Checking Normality by Eye, With a Worked Example

Exploratory data analysis and normal distribution both mention Q-Q plots as a way to check normality visually, without walking through how one is actually built. A Q-Q plot ("quantile-quantile plot") compares a dataset's own quantiles against the quantiles a theoretical distribution would predict — when the two line up, the points fall on a straight diagonal line; when they don't, the departure shows up as a specific, readable curve.

How a Q-Q Plot Is Built

For each data point, a Q-Q plot pairs two values: the data point's own rank-based position in the sorted sample (its "sample quantile"), and the value a normal distribution would predict at that same relative position (its "theoretical quantile"). Plotting sample quantiles against theoretical quantiles, point by point, produces the Q-Q plot. If the sample genuinely came from a normal distribution, sample and theoretical quantiles should track each other closely, and the points fall close to a straight diagonal line. This makes a Q-Q plot considerably more sensitive than a histogram for spotting subtle departures, especially in a distribution's tails, where a histogram's bars can visually smooth over exactly the detail a Q-Q plot exposes.

Worked Example

Problem: A small dataset of 6 exam scores: 62, 68, 71, 75, 78, 95. Construct the theoretical quantiles needed for a normal Q-Q plot, and assess whether the data looks normally distributed.

Step 1: Sort the data (already sorted): 62, 68, 71, 75, 78, 95.

Step 2: Find each point's plotting position, a common convention being (i−0.5)/n for the i-th sorted value out of n total: for n=6, these are 0.083, 0.25, 0.417, 0.583, 0.75, 0.917.

Step 3: Convert each plotting position to a standard normal z-quantile (the z-score at which the standard normal CDF equals that probability): −1.383, −0.674, −0.210, 0.210, 0.674, 1.383.

Step 4: Convert theoretical z-quantiles into the data's own units, using the sample's mean (74.83) and standard deviation (11.34), so the comparison is on the same scale as the actual data: 59.15, 67.19, 72.45, 77.22, 82.48, 90.51.

Comparing sorted actual data to these theoretical values pair by pair: 62 vs. 59.15 (close), 68 vs. 67.19 (close), 71 vs. 72.45 (close), 75 vs. 77.22 (close), 78 vs. 82.48 (starting to drift), and 95 vs. 90.51 (the clearest departure — the actual value sits well above what a normal distribution would predict at that position). Plotted, the first five points would sit fairly close to the diagonal line, with the sixth point pulling noticeably above it — the visual signature of a single high outlier or a right-skewed tail, exactly what the raw data (with one score of 95 well above the rest) would suggest.

Reading Common Departure Patterns

Points curving upward at the right, downward at the left (an S-shape) — heavier tails than a normal distribution: more extreme values on both ends than the normal model predicts.

Points curving in one consistent direction throughout — skewness: a systematic bend across the whole plot (not just at the ends) generally signals a right-skewed or left-skewed distribution rather than a genuinely normal one.

Most points on the line, with one or two clear outliers — exactly the pattern in the worked example above: a data point that jumps noticeably off the line at one end, while the bulk of the data tracks the line closely, suggests a small number of extreme values rather than a fundamentally different overall shape.

Why Pair a Q-Q Plot With a Formal Test

A Q-Q plot is a visual, judgment-based tool — "close enough to a line" is inherently somewhat subjective, especially with a small sample like the one above. Pairing it with a formal normality test (like the Shapiro-Wilk test, which produces a numeric verdict) gives a quantified check alongside the visual impression, exactly the combined approach recommended in exploratory data analysis. The two serve different, complementary purposes: the Q-Q plot shows how and where a distribution departs from normality (which a single test statistic doesn't reveal), while the formal test gives an objective, quantified answer to whether the departure is likely to matter statistically.

Real-World Uses

Checking regression residuals: a Q-Q plot of a regression model's residuals is one of the standard diagnostic checks recommended in simple linear regression — since the model's inferential guarantees rely on approximately normal residuals, a Q-Q plot is often the first place a violation becomes visible.

Financial return analysis: asset returns are famously heavier-tailed than a normal distribution predicts, and a Q-Q plot's characteristic S-curve is a standard, quick way analysts visualize and communicate exactly how much a return series departs from normality before applying models that assume it.

Choosing between parametric and non-parametric methods: a Q-Q plot showing a clear, systematic departure from the diagonal line is often the deciding visual evidence for switching from a parametric test to one of the non-parametric alternatives covered elsewhere on this site.

Common Mistakes

Expecting a perfectly straight line from real data. Even genuinely normal data shows some random wiggle around the line, especially with a small sample — the question is whether the pattern looks like random noise around a line, or a systematic, structured curve.

Judging normality from a Q-Q plot with very few data points. With only a handful of observations (as in the worked example above), even genuinely normal data can produce a somewhat irregular-looking plot purely by chance — larger samples give a much more reliable visual signal.