What Random Scatter Looks Like
In “Constructing a Residual Plot,” you learned to plot each residual vertically against its predictor \(x\) or fitted value \(\hat{y}\), with a horizontal reference line at zero. The next step is to read the plot. For a linear model, a useful residual plot has points scattered on both sides of zero without a clear, systematic pattern.
“Random scatter” is a visual description, not a claim that the observations were randomly assigned or that every residual was generated by a particular probability model. It means that, as you move across the horizontal axis, the residuals do not show a noticeable curve, trend, or other organized shape. The points should generally remain centered around zero across the horizontal range, rather than being mostly positive in one region and mostly negative in another.
A plot that looks unstructured does not need to have exactly the same number of points above and below zero, or points at equal distances from the line. Instead, look for an overall balance across the horizontal range. The typical vertical distance from zero should also be reasonably similar across that range; a pronounced change in spread can be a pattern worth investigating.
Read the Plot Across Its Range
The least-squares line has residuals that sum to zero, as discussed in “Why Residuals Sum to Zero.” Consequently, the average of all the residuals is zero. But that arithmetic fact does not establish that the residual plot is unstructured. Residuals can balance overall while forming a noticeable shape. The visual question is whether they remain scattered around zero as the horizontal variable changes.
Use the horizontal reference line to see which residuals are positive and which are negative.
Notice whether residuals remain on both sides of zero, or whether they form a run, trend, or other organized shape across the horizontal range.
Check whether residuals stay at a broadly similar distance from zero or whether the spread changes substantially across the plot.
If the plot appears unstructured and centered around zero, say that it supports using a linear model for these data over the observed range. Do not claim that the plot proves the model is perfect.
A residual plot is a diagnostic for the form of the relationship, not a score for how strong the relationship is. A plot can look unstructured even when residuals are fairly far from zero, and a plot can have small residuals while still showing a pattern. Judge the pattern itself; do not use the size of the residuals alone as a substitute for that judgment.
Worked Example: Scatter Around Zero for a Greenhouse Model
A fictional greenhouse team models the amount of water used by a test bed, \(y\) liters, from the number of days since setup, \(x\). The fitted line is \(\hat{y}=10+2x\). The residual plot against \(x\) has these coordinates:
| \(x\) (days) | \(y\) (liters) | \(\hat{y}\) (liters) | Residual \(y-\hat{y}\) (liters) |
|---|---|---|---|
| 1 | 13 | 12 | 1 |
| 2 | 12 | 14 | −2 |
| 3 | 18 | 16 | 2 |
| 4 | 18 | 18 | 0 |
| 5 | 19 | 20 | −1 |
| 6 | 23 | 22 | 1 |
| 7 | 20 | 24 | −4 |
| 8 | 29 | 26 | 3 |
State. The question is whether the residuals show a clear pattern against days since setup, and whether their appearance supports a linear model.
Plan. Check the residual signs and sizes as \(x\) increases. Also verify that the entries are consistent with the fitted line and residual definition from “Defining a Residual as Observed Minus Predicted.”
Do. At \(x=1\), the prediction is \(10+2(1)=12\) liters, so the residual is \(13-12=1\) liter. At \(x=7\), the prediction is \(10+2(7)=24\) liters, so the residual is \(20-24=-4\) liters. Applying the same calculation to each row gives the residual sequence \(1,-2,2,0,-1,1,-4,3\) liters. The points fall both above and below zero through the horizontal range; they do not steadily rise or fall, and they do not trace an obvious curve.
As a numerical check, the residuals sum to \(1-2+2+0-1+1-4+3=0\) liters. To check their balance with the horizontal coordinates, the products of \(x\) and residual are \(1,-4,6,0,-5,6,-28,24\), which sum to zero. These checks are consistent with residuals from a least-squares line, but the visual reading still matters: inspect the plotted points rather than treating a zero sum as proof that the plot has no pattern.
Conclude. The residual plot appears unstructured and scattered around zero. This supports using a linear model for water use and days since setup over the observed range, although it does not prove that the model describes every aspect of the relationship.
Overall Balance Is Not Enough
Because the residuals from a least-squares line with an intercept sum to zero, they can be balanced overall even when the plot has an obvious pattern. Imagine negative residuals near the middle of the horizontal range and positive residuals at both ends. The positive and negative values might cancel in the total, yet their arrangement would not look like unstructured scatter.
This is why you should read the order and location of the points, not just calculate a mean residual. A practical visual check is to mentally divide the horizontal axis into a few regions. Ask whether residuals in each region still appear reasonably centered around zero. This is not a formal calculation or a rule requiring equal counts; it is a way to notice when one region sits mostly above or below the reference line.
Worked Example: Residuals Can Sum to Zero and Still Show a Pattern
A fictional parks team uses a linear model to predict the minutes needed to complete a short trail loop from the number of walkers in a group, \(x\). Across six observed group sizes, a residual plot against \(x\) shows these residuals, in minutes, from left to right:
| Group-size position from left to right | Residual (minutes) |
|---|---|
| 1 | 3 |
| 2 | −1 |
| 3 | −2 |
| 4 | −2 |
| 5 | −1 |
| 6 | 3 |
State. Decide whether these residuals form unstructured scatter around zero, even though they balance overall.
Plan. First check their sum. Then consider their sequence across the horizontal axis, because a zero sum by itself does not show whether they are arranged without a pattern.
Do. Their sum is \(3-1-2-2-1+3=0\) minutes. Nevertheless, the plotted sequence starts above zero, lies below zero through the middle, and returns above zero at the right. That visible arrangement is not random-looking scatter around the reference line. The values are also symmetric from the two ends toward the center, making the shape especially apparent.
Conclude. These residuals are centered at zero overall but not unstructured. The plot does not provide the same support for a linear model as a plot with no clear pattern. The key distinction is the residuals’ arrangement across group size, not just their total.
What This Diagnostic Can and Cannot Tell You
A residual plot with unstructured scatter centered around zero gives support for the linearity of a model over the range shown. In plain language, the plot does not reveal a clear way in which the fitted line systematically misses the observations as the predictor changes. That is useful evidence when deciding whether a straight-line description is reasonable.
The conclusion should stay within that scope. A residual plot does not prove that a linear model is correct, establish that the association is strong, or show that one variable causes the other. It also does not, by itself, settle every other question about the data or the model. Describe what the plot shows and use cautious language such as “supports using a linear model” rather than “proves the relationship is linear.”
Keep the horizontal axis in mind as you interpret the result. A residual plot against \(x\) describes how residuals behave across predictor values. A plot against \(\hat{y}\) describes how they behave across predicted responses. As in “Constructing a Residual Plot,” each residual must stay paired with the correct horizontal coordinate.
Worked Example: Writing a Careful Interpretation
A fictional school technology team predicts the number of minutes a classroom device takes to finish a diagnostic, \(y\), from the size of the file being checked, \(x\). A residual plot against file size has points distributed above and below zero throughout the observed range. No clear trend or curve is visible, and the vertical spread looks broadly similar from smaller to larger files.
State. Decide what this pattern suggests about using a linear model for diagnostic time and file size.
Plan. Describe the visible residual pattern in context, then give a conclusion limited to what the residual plot can support.
Do. The residuals appear scattered around zero across the observed file sizes, with no obvious systematic pattern and no pronounced change in vertical spread. The plot therefore does not show a clear reason, based on its pattern, to reject a straight-line description for this range.
Conclude. The residual plot supports using a linear model to describe the relationship between file size and diagnostic time over the file sizes observed. It does not prove that the relationship is exactly linear or that the model will be appropriate outside that range.
Common Mistakes and AP Exam Tips
- Calling any zero-sum residual set random scatter. A least-squares line with an intercept has residuals that sum to zero, but the points can still form a visible pattern. Describe how they are arranged across the horizontal axis.
- Looking only at whether there are positive and negative residuals. Both signs may occur while one section of the plot is mostly above zero and another is mostly below. Scan from left to right.
- Requiring perfect symmetry. An unstructured plot does not need matching points or exactly equal numbers above and below zero. Look for the absence of an obvious systematic pattern, not a perfect design.
- Saying the model is proven correct. A plot with random-looking scatter supports a linear model; it does not prove the model is perfect or establish causation.
- Ignoring the observed range. Describe the conclusion for the predictor values shown in the plot. Do not extend the conclusion beyond those values without additional evidence.
- Equating small residuals with unstructured residuals. A plot can have small residuals arranged in a noticeable pattern. Assess the shape and balance, not just the distances from zero.
For a full-credit interpretation, identify the residuals as scattered around zero, state whether a clear pattern is present, and connect that observation to the appropriateness of a linear model in context. Prefer “This supports using a linear model for these data over the observed range” to an absolute claim that the relationship is linear.
Check Your Understanding
Use the pattern across the horizontal range—not just the residual sum—to answer these questions.
- What does it mean for residuals to be scattered around zero without a clear pattern?
- Why does a residual sum of zero fail to show, by itself, that a plot has unstructured scatter?
- A residual plot has mostly negative values on the left and mostly positive values on the right. Does that look unstructured? Explain briefly.
- What conclusion is appropriately supported by an unstructured residual plot centered around zero?
- Why should a conclusion about linearity refer to the observed range?