Outlier vs high-leverage point in regression
By Jude Wallis · Published
In regression, an outlier is a point that falls far from the pattern of the other points in the vertical direction, so it has a large residual. A high-leverage point has an x-value far from the mean of x. A point can be one, both, or neither, and either kind may or may not change the line.
AP Statistics: Unit 5 (topics 5.1 Graphical Representations Between Two Quantitative Variables, 5.4 Residuals). Topic 5.1 of the Fall 2026 AP Statistics course asks you to describe unusual features of a scatterplot, such as points that do not fit the general pattern, and topic 5.4 covers residuals. The terms high leverage and influential point are standard statistics vocabulary for making that description precise; treat them as general statistics background and not as framework wording.
Outlier vs high-leverage point: the short answer
The two labels describe a point's position in two different directions on a scatterplot.
An outlier in regression is unusual vertically. It sits well above or below where the pattern of the other points says it should be, so its residual, observed minus predicted , is large in size.
A high-leverage point is unusual horizontally. Its -value is far from ("x-bar", the mean of the -values), out to the left or right of the rest of the data. Leverage depends only on , so you can spot it without looking at .
A third label describes effect instead of position. An influential point is one whose removal changes the slope, the intercept, or the correlation substantially. Outliers and high-leverage points are both candidates for being influential, and you find out by fitting the line with and without the point.
What an outlier is in regression
With one variable, an outlier is a value far from the rest of a distribution, often flagged with the 1.5 IQR rule. With two variables the idea changes. A regression outlier is a point that does not follow the general pattern of the scatterplot and has a large residual.
The point's -value does not have to be extreme. If most students who study 6 hours score near 15 on a quiz and one scores 30, that point is an outlier in the regression even if scores of 30 occur elsewhere in the data for students who studied longer. It is unusual for its .
An outlier whose is near has little power to tilt the line, because the line pivots around the middle of the data. It mostly pulls the whole line up or down slightly, weakens the correlation, and increases the typical size of the residuals.
What a high-leverage point is
A high-leverage point has an -value far from the other -values. If five houses have between 1,000 and 2,000 square feet and a sixth has 6,000, the sixth has high leverage whatever its price.
The name comes from a lever. The least-squares line always passes through , and a point far from sits at the long end of the lever, where a change in its -value swings the slope the most.
High leverage is potential. If the point's -value lies along the trend set by the other points, the line barely moves and the point just extends the pattern. If its -value is off that trend, the line tilts toward it. In the second case the tilting can make the point's own residual small, so a point that has changed the line a great deal may not look like an outlier on the residual plot.
The differences side by side
| Feature | Outlier (in regression) | High-leverage point |
|---|---|---|
| Unusual in which direction | Vertical: far from the line | Horizontal: far from |
| Identified by | A large residual | An extreme -value |
| Depends on | Yes | No |
| Where to look | Scatterplot or residual plot | Scatterplot, along the horizontal axis |
| Typical effect when not also the other kind | Shifts the line up or down, lowers the size of | Little change to the line if it follows the trend |
| Residual | Large | Can be small, even when the point has moved the line |
| Automatically influential | No | No |
A point can carry both labels. A point with an extreme that is also far from the trend of the others is the most likely kind to be influential.
How to tell which one you have
Work through three checks.
- Look along the horizontal axis. If the point's -value is separated from the rest, it has high leverage.
- Look at the vertical distance from the pattern of the other points. If that distance is large compared with the other points' distances, it is an outlier.
- Fit the line with the point and without it. If the slope, intercept, or correlation changes noticeably, the point is influential. Say which of those changed.
The third check is the only way to settle influence, and the regression influential point interactive lets you drag a point and watch the line respond.
Finding an unusual point is a reason to investigate it. Check for a recording error and consider whether the point belongs to the same population as the others. Removing a point requires a reason beyond its being inconvenient. When in doubt, report the results both ways.
The classic mix-up and how to avoid it
The first error is assuming that every unusual point is influential. An outlier in the middle of the -values may leave the slope unchanged. A high-leverage point that follows the trend may leave the whole line unchanged. Position does not settle influence.
The second error is using the residual plot as the only detector. A high-leverage point pulls the line toward itself, which shrinks its own residual and enlarges the residuals of other points. In the second worked example below, the point that changes the slope the most has only the third-largest residual of the six. Look at the scatterplot as well.
The third error is carrying over the one-variable meaning of outlier. In regression, a point with ordinary and ordinary values can still be an outlier if the combination is far from the pattern.
In written answers, describe what you see: the point's -value is far from the others, or the point lies far from the line. Then state what it does to the slope, the intercept, or the correlation.
An outlier at the mean of x
Five points, (2, 7), (4, 10), (6, 16), (8, 17), and (10, 25), have least-squares line and correlation . A sixth point, (6, 30), is added. The new least-squares line is with . Classify the new point and describe its effect.
Check the horizontal position. The original -values have mean , and the new point has . It sits exactly at the mean, so it does not have high leverage.
Check the vertical position against the original line. The prediction at is , and the observed value is 30, which is 15 above the line. The other five points are all within 2.3 of that line. The point is an outlier.
Find its residual from the new line: the prediction is , so the residual is . The other five residuals from the new line are -1.9, -3.2, -1.5, -4.8, and -1.1, so 12.5 is by far the largest in size.
Compare the slopes: 2.15 before and 2.15 after. The slope did not change at all.
Compare the intercepts: 2.1 before and 4.6 after. The line moved up by 2.5 at every , which is the outlier's 15-unit gap spread over 6 points, .
Compare the correlations: fell from about 0.976 to about 0.696, because the points are now more scattered around the line.
The point (6, 30) is an outlier with a residual of 12.5, and it is not a high-leverage point because its -value equals the mean. It leaves the slope at 2.15, raises the intercept from 2.1 to 4.6, and lowers the correlation from about 0.976 to about 0.696. It is influential for the intercept and the correlation, with no effect on the slope.
Two high-leverage points with different effects
Start again with the five points (2, 7), (4, 10), (6, 16), (8, 17), (10, 25) and the line , . Consider adding one of two points at . Adding (20, 45.1) gives the line with . Adding (20, 20) gives with . Classify each added point.
Check the horizontal position. The other -values run from 2 to 10, and is far to the right of all of them. Both candidate points have high leverage.
Check (20, 45.1) against the original line: . The point lies exactly on the line, so its residual is 0 and it is not an outlier.
State its effect. The slope and intercept are unchanged. The correlation rises from about 0.976 to about 0.995, because the point extends the linear pattern over a wider range of .
Check (20, 20) against the original line: it is below the prediction, far from the trend of the other five points.
State its effect. The slope falls from 2.15 to about 0.71, the intercept rises from 2.1 to about 9.92, and the correlation falls to about 0.691. This point is influential.
Find its residual from the new line: using less rounded coefficients, the prediction is , so the residual is about . The residuals of the other five points are about -4.34, -2.76, 1.82, 1.40, and 7.98. The added point's residual is only the third largest in size, because the line swung toward it.
Both points have high leverage because is far from the other -values. The point (20, 45.1) follows the trend: it is not an outlier, it leaves the line unchanged, and it raises to about 0.995. The point (20, 20) departs from the trend and is highly influential, cutting the slope from 2.15 to about 0.71, yet its residual of about -4.11 is smaller in size than the 7.98 residual at (10, 25).
Frequently asked questions
Can a point be both an outlier and a high-leverage point?
Yes. A point with an -value far from the others that also lies far from the trend of the other points is both. Such a point usually has a strong effect on the slope, and its residual from the fitted line may be modest because the line has been pulled toward it.
Is a high-leverage point always influential?
No. If its -value falls along the trend of the other points, removing it leaves the slope and intercept nearly unchanged. It may still raise the size of the correlation. Leverage is the potential to change the line, and influence is the change that occurs.
Should I remove an outlier or high-leverage point from the data?
Only with a reason you can state, such as a confirmed recording error or a case that does not belong to the population being studied. Otherwise keep it, and if it changes the conclusions, report the regression with and without the point.