Normal Probability Plots
Normal Probability Plots: Checking the Normality Assumption
A normal probability plot (also called a Q‑Q plot) compares the ordered values of a sample with the values you would expect if the sample came from a normal distribution. If the sample really is normal, the points fall close to a straight line; systematic curvature is what signals a departure from normality.
Building the plot
Order the data x(1)≤x(2)≤⋯≤x(n). Assign each observation a plotting position (an estimate of its percentile), most commonly
Convert that proportion to a theoretical z‑score with the inverse standard normal function,
and plot the pairs (zi,x(i)).
Throughout this concept the theoretical z‑scores are on the horizontal axis and the ordered data are on the vertical axis.
Software packages differ here, some put the data on the horizontal axis, or label the horizontal axis in percent rather than in z, so always check the axes before reading a curve's direction.
Worked mini-example (n = 5)
i | x(i) | (i−0.5)/n | zi |
|---|---|---|---|
1 | 70 | 0.10 | -1.28 |
2 | 74 | 0.30 | -0.52 |
3 | 78 | 0.50 | 0.00 |
4 | 82 | 0.70 | 0.52 |
5 | 90 | 0.90 | 1.28 |
Plotting these five (zi,x(i)) pairs and drawing a reference line through the middle of the cloud lets you judge linearity by eye. Small, roughly linear scatter supports normality.
Reading the shape
Once the axes are fixed as above, the departures have standard meanings. An upper-tail departure above the reference line at high z suggests right skew when the lower and middle points remain close to the line.
If points instead depart outward at both ends, below the line at low z and above it at high z, the pattern suggests both tails are heavier than normal.
The mirror image of the one-sided pattern, points bowing below the line at low z only, suggests left skew.
When those departures are approximately mirror images, they are consistent with a symmetric heavy-tailed distribution; outward departures at both ends alone do not establish symmetry. A two‑sided inward bow says both tails are lighter than normal.
Normal probability plot of the ordered ages of 10 patients against the theoretical z-scores from the (i - 0.5)/n convention. The dashed line is a reference line fitted through the lower and middle points.
The two visuals for this concept show the contrast. The battery‑discharge sample in the chart stays close to a straight line across its whole range, which is what a sample drawn from a normal population looks like.
The patient‑age sample in the figure tracks its reference line at the low and middle z values and then departs from it near the top.
Normal probability plot for a sample of 12 battery discharge times
A statistician is constructing a normal probability plot for a sample of n=15 observations, using the plotting-position convention ni−0.5. What plotting position (cumulative proportion) should be assigned to the 4th smallest observation?
Give your answer in proportion (unitless).
The figure shows a normal probability plot of ages (in years) for a sample of 10 patients, with a reference line fitted through the bulk of the data. The two largest ages fall noticeably above the reference line, while the smaller ages track the line closely. What does this pattern suggest about the age distribution?
Normal probability plot of the ordered ages of 10 patients against the theoretical z-scores from the (i - 0.5)/n convention. The dashed line is a reference line fitted through the lower and middle points.
Normal probability plot for a sample of 12 battery discharge times