From a Residual Calculation to a Full Response
In “Common Residual Mistakes,” you practiced checking the subtraction order, sign, units, and interpretation of an individual residual. This tutorial puts those skills together in an AP-style free-response question and adds a careful reading of the residual plot. A complete response does more than report a number: it shows how the prediction was found, explains what the residual means for the case, and describes any pattern in the plot.
For an observed case, the residual is \(y-\hat{y}\), where \(y\) is the observed response and \(\hat{y}\) is the response predicted by the fitted line at that case’s predictor value. As in “Sign of a Residual: Over- and Underprediction,” a positive residual means the observation is above the prediction, so the line underpredicted. A negative residual means the observation is below the prediction, so the line overpredicted.
A useful residual plot reading has two parts. First, describe what the residuals do across the horizontal range: do they scatter above and below zero without a clear pattern, or do they form a curve, fan, or another organized arrangement? Second, explain what that pattern suggests about the linear model. Random-looking scatter supports using a linear model for the data; an organized pattern indicates that the line misses systematically in some way.
Keep the question’s variables and units in view throughout. The horizontal values may be predictor values or fitted responses, but residuals always use the response variable’s units. Also, distinguish one residual from the overall plot: a single positive residual describes one case, while a pattern describes how the model’s errors behave across many cases.
A Reliable Free-Response Structure
For a question that asks you to calculate and interpret a residual, use the fitted equation at the case’s observed \(x\), then subtract the prediction from the observed response. Write the formula before substituting so the order is clear. For a plot-reading part, refer to the zero line and describe the pattern over the range rather than commenting on a single point alone.
Identify the case or model feature the question asks about. Name the response and its units when calculating a residual.
For a residual, substitute the case’s \(x\)-value into the fitted line, then use \(y-\hat{y}\). For a residual plot, compare the vertical positions with zero and look across the horizontal range for structure.
Show the prediction and subtraction. For a plot, describe the pattern accurately, such as “scattered above and below zero without a clear curve.”
Interpret the residual in context, or connect the plot’s pattern to whether the linear model appears reasonable for the data.
This structure is not a substitute for answering the exact question. If asked whether the model overpredicted for one case, say so explicitly. If asked what a plot suggests, explain whether the pattern supports a linear model or signals a systematic miss. Avoid a vague conclusion such as “the graph looks good.”
Worked Example: Calculate and Interpret One Residual
A fictional greenhouse uses daily hours of supplemental light \(x\) to predict the height of a young plant \(y\), in centimeters. The fitted line is \(\hat{y}=7.2+1.8x\). One plant received 6 hours of supplemental light and grew to 18.4 cm. Calculate and interpret its residual.
State. Find the residual for this plant and explain how its observed height compares with the model’s prediction.
Plan. Plant height is the response, so the residual is measured in centimeters. Substitute \(x=6\) into the fitted line to find the predicted height, then calculate observed height minus predicted height and use the sign to describe the difference.
Do. The fitted line predicts \(\hat{y}=7.2+1.8(6)=7.2+10.8=18.0\) cm. Therefore, \(y-\hat{y}=18.4-18.0=0.4\) cm. The observed height is greater than the predicted height, which agrees with the positive residual.
Conclude. This plant’s residual is \(0.4\) cm. Its observed height was 0.4 cm greater than predicted, so the model underpredicted this plant’s height by 0.4 cm.
Reading Scatter Without a Clear Pattern
A residual plot supports a linear model when the residuals are scattered above and below zero without a clear systematic pattern across the horizontal range. “Scattered” does not mean that every residual must be close to zero or that the points must be evenly spaced. Individual residuals can be relatively large; what matters when judging the form of the relationship is whether their arrangement reveals a consistent miss, such as a curve.
Describe what is visible rather than making an absolute claim. For example, “The residuals vary above and below zero with no clear curved pattern, so the plot supports using a linear model” is more informative than “the residuals are random.” A residual plot does not prove that a linear model is correct; it provides evidence about whether the line’s errors show a recognizable pattern.
Worked Example: Interpreting a Residual Plot with No Clear Curve
A fictional sports program models a player’s free-throw success rate \(y\), in percentage points, from minutes of weekly practice \(x\). Its residual plot uses \(x\) horizontally. The plotted coordinates are shown below; each pair is \((x,\text{residual})\).
| Practice minutes, \(x\) | Residual (percentage points) |
|---|---|
| 10 | 1 |
| 15 | -2 |
| 20 | 1 |
| 25 | 1 |
| 30 | -2 |
| 35 | 1 |
| 40 | 0 |
Describe the plot and state what it suggests about using a linear model.
State. Assess whether the residuals show a systematic pattern across practice times and explain what that means for the linear model.
Plan. Compare the residuals with zero and look across the \(x\)-values for a curve or other consistent arrangement. Interpret their vertical values in percentage points, the response variable’s units.
Do. The residuals are sometimes above and sometimes below zero over the full practice-time range. They do not form a clear curve or a steadily widening or narrowing spread. There are seven observations, so the plot cannot establish that the model is perfect, but it does not show an obvious systematic pattern.
Conclude. The residuals are scattered above and below zero without a clear pattern, which supports using a linear model for these data. This conclusion concerns the model’s residual pattern; it does not prove that practice time causes changes in free-throw success.
The listed residuals are also consistent with a least-squares line that includes an intercept. Their sum is \(1-2+1+1-2+1+0=0\), and the predictor-weighted sum is \(10(1)+15(-2)+20(1)+25(1)+30(-2)+35(1)+40(0)=0\). As discussed in “Properties of the Least-Squares Line,” these are expected properties of the fitted residuals. Checking them is a useful safeguard when constructing or reviewing a set of claimed least-squares residuals.
When the Residual Plot Shows a Curve
A curved pattern means that residuals are not merely scattered around zero: their positions change in an organized way across the horizontal range. As in “Curved Patterns in a Residual Plot,” a U-shaped arrangement often has positive residuals at both ends and negative residuals in the middle, or the reverse. Positive residuals mean the observed responses are above the line; negative residuals mean they are below it. The line is therefore making errors in a systematic direction at different parts of the range.
When explaining a curve, state its shape, where the residuals are positive or negative, and what that says about predictions. Then conclude that the residual pattern suggests a straight-line model does not capture the relationship well. Do not claim that every individual prediction is poor, and do not name a specific alternative model unless the question asks for one.
Worked Example: Explain a U-Shaped Residual Pattern
A fictional greenhouse models daily electricity use \(y\), in kilowatt-hours, from the outside temperature \(x\), in degrees Celsius. A residual plot from its fitted least-squares line has these coordinates:
| Temperature, \(x\) (°C) | Residual (kWh) |
|---|---|
| 12 | 5 |
| 14 | 0 |
| 16 | -3 |
| 18 | -4 |
| 20 | -3 |
| 22 | 0 |
| 24 | 5 |
Describe the residual pattern and explain what it indicates about the line’s predictions.
State. Decide whether this residual plot supports a linear model for predicting electricity use from temperature.
Plan. Describe the residuals’ arrangement relative to zero across the temperature range. Translate positive and negative residuals into underprediction and overprediction in kilowatt-hours.
Do. Residuals are positive at the low and high temperatures, near zero around the transitions, and negative in the middle. This is a U-shaped pattern. At the ends, the observed electricity use is above the line’s predictions, so the line underpredicts. In the middle, observed use is below the predictions, so the line overpredicts. The residuals sum to \(5+0-3-4-3+0+5=0\); their predictor-weighted sum is \(12(5)+14(0)+16(-3)+18(-4)+20(-3)+22(0)+24(5)=0\), consistent with least-squares residual properties.
Conclude. The residual plot shows a clear U-shaped pattern, so the straight-line model misses systematically and does not capture the relationship between temperature and electricity use well. It underpredicts at both ends of the temperature range and overpredicts in the middle.
Common Mistakes and AP Exam Tips
- Giving a residual without showing the prediction. A full calculation shows the fitted line evaluated at the case’s \(x\), followed by \(y-\hat{y}\). This makes it possible to check the subtraction and sign.
- Reversing observed and predicted. The residual is observed minus predicted, not predicted minus observed. Check that a positive result matches an observed response above the prediction.
- Using the predictor’s units. Residuals have the response variable’s units. If the response is free-throw percentage, for example, describe its residual in percentage points rather than practice minutes.
- Describing only the zero line. Saying “the residuals are around zero” is incomplete if the plot has a clear curve or changing spread. Describe the arrangement across the full horizontal range.
- Calling every plot random or perfect. Prefer a qualified description such as “no clear systematic pattern is visible.” A residual plot can support a linear model, but it does not prove the model is correct.
- Reading a curve backwards. Positive residuals mean underprediction; negative residuals mean overprediction. State both the plot’s shape and the direction of the errors in context.
A strong AP response uses precise language tied to the question. For a residual calculation, show the prediction, calculate observed minus predicted, include the response units, and explain whether the line overpredicted or underpredicted. For a plot, describe the pattern and connect it to the linear model. Avoid relying on phrases such as “the data are good” or “the line fits” without explaining what the residuals show.
Check Your Understanding
Answer each question with the calculation or plot description requested, including context and units where appropriate.
- A model predicts a runner’s 5-kilometer time to be 26.5 minutes. The runner’s observed time is 25.8 minutes. Calculate and interpret the residual.
- A fitted line is \(\hat{y}=4+2.5x\). For an observed case with \(x=3\), the response is \(y=12.2\). Find the residual and state whether the line overpredicted or underpredicted.
- A residual plot has points on both sides of zero, but the vertical spread grows steadily as the horizontal value increases. Describe the pattern and what it suggests.
- A residual plot is U-shaped, with positive residuals at both ends and negative residuals in the middle. Explain the line’s prediction errors across the range.
- Why is “the residuals are close to zero” not enough to describe a plot if the residuals form a clear curve?