A Q-Q plot, or quantile-quantile plot, is a graph used to compare the shape of a data set to a chosen theoretical distribution. For normality checks, the sample data quantiles are plotted against the quantiles expected from a normal distribution. If the points fall close to a straight reference line, the data are reasonably consistent with that distribution.
This matters because many statistical methods assume normality or use normal models as an approximation.
Each point on a Q-Q plot matches a ranked data value with the theoretical quantile at the same cumulative probability. The diagonal reference line shows where the points would fall if the sample followed the theoretical distribution exactly, apart from sampling variation. Curved patterns, S-shapes, or extreme points far from the line reveal differences in skewness, tail behavior, or outliers.
Q-Q plots are especially useful because they show not only whether a distribution differs from normal, but also how it differs.
Understanding Statistics: Q-Q Plots
A normal Q-Q plot is built after the data values have been sorted from smallest to largest. Sorting is essential because quantiles describe positions within the ordered list. The smallest observations are compared with the lower end of a normal distribution, middle observations with its centre, and the largest observations with its upper end.
A sample never matches a model perfectly. Random variation makes points wobble around the line, especially in a small sample. The useful skill is judging the overall pattern rather than treating every small gap as evidence of failure.
The direction of a curve gives clues about the data shape. Right skewed data often bend upward at the high end because a few large values stretch far beyond what a normal model expects. Left skewed data show the opposite tendency near the low end.
Data with heavy tails tend to move away from the line at both ends. This means unusually extreme values occur more often than under a normal model.
Light tailed data can bend inward at both ends because their values stay closer to the centre. An S shaped pattern can indicate that the middle and tails have different spreads from a normal distribution.
One far away point deserves careful attention, but it does not automatically mean the observation is wrong. It may be a recording error, a genuine unusual event, or a sign that several groups were mixed together. For example, test scores from two classes with very different teaching conditions may form a pattern that does not fit one normal distribution.
Heights of adults and children mixed in one data set can do the same. Before removing any value, students should check how it was measured and whether it belongs to the population being studied. Removing inconvenient results without a reason can give a misleading conclusion.
Q-Q plots appear when students use methods such as confidence intervals, t tests, regression, and analysis of variance. In regression, the plot is usually made from residuals rather than from the original response values. Residuals are the differences between observed values and model predictions.
A normal pattern in residuals supports some common calculations for uncertainty. Yet normality is only one condition to check. Independence, equal spread, sensible sampling, and the absence of strong groups can matter just as much.
With large data sets, tiny harmless departures can look clear on a plot. With very small data sets, serious departures can be hard to see. Students should combine the graph with knowledge of the context, sample size, and the purpose of the analysis.
Key Facts
- A Q-Q plot compares sample quantiles to theoretical quantiles from a chosen distribution.
- For a normal Q-Q plot, the x-axis is theoretical normal quantiles and the y-axis is ordered sample data values.
- If the data are approximately normal, the plotted points should lie close to a straight reference line.
- A common plotting position is p_i = (i - 0.5) / n for the i-th ordered value in a sample of size n.
- The theoretical quantile is x_i = F^-1(p_i), where F^-1 is the inverse cumulative distribution function.
- Systematic curvature away from the line indicates a distributional difference such as skewness, heavy tails, or light tails.
Vocabulary
- Quantile
- A quantile is a value that divides a distribution so that a given fraction of data falls at or below it.
- Q-Q Plot
- A Q-Q plot is a graph that compares the quantiles of one data set to the quantiles of another data set or theoretical distribution.
- Theoretical Quantile
- A theoretical quantile is the value predicted by a probability distribution at a specified cumulative probability.
- Reference Line
- The reference line is the straight line that represents the expected pattern if the sample follows the theoretical distribution.
- Normality
- Normality means that data follow a bell-shaped normal distribution, or are close enough for a normal model to be useful.
Common Mistakes to Avoid
- Expecting every point to lie exactly on the line. Real samples have random variation, so small departures from the reference line are normal.
- Judging normality from one or two extreme points only. Isolated outliers matter, but the overall pattern of the points gives the strongest evidence.
- Ignoring the direction of curvature. An S-shaped pattern, upward bend, or downward bend can indicate different issues such as skewness or heavy tails.
- Using a normal Q-Q plot without checking whether a normal model is appropriate for the variable. Counts, proportions, and strongly bounded measurements may need a different distribution or transformation.
Practice Questions
- 1 A data set has n = 20 observations. Using p_i = (i - 0.5) / n, find the plotting position for the 5th ordered value.
- 2 In a normal Q-Q plot, the 10th ordered sample value is 12.4 and its theoretical normal quantile is 0.85. If the fitted reference line is y = 10 + 3x, what value does the line predict at x = 0.85, and is the sample point above or below the line?
- 3 A normal Q-Q plot has points close to the line in the middle but far above the line on the right end and far below the line on the left end. What does this pattern suggest about the tails of the data compared with a normal distribution?