Common Residual Mistakes

Why Residuals Are Easy to Misread

In “Reading Residuals From Computer Output,” you learned to find a residual in a table or list and match it to the correct observation. Once you have the number, you still need to interpret it accurately. Three mistakes are especially common: subtracting in the wrong order, omitting the response units, and treating the sign as if it described something other than the observed response’s position relative to the prediction.

Recall from “Defining a Residual as Observed Minus Predicted” that a residual for an observed case is \(y-\hat{y}\). Here, \(y\) is the observed response and \(\hat{y}\) is the response predicted by the fitted line at that case’s predictor value. The order is part of the definition. A residual is not \(\hat{y}-y\).

The sign does not tell you whether the predictor is high or low, whether the fitted line has a positive or negative slope, or whether the observation is a regression outlier. It tells you whether the observed response is greater than or less than its prediction. As in “Sign of a Residual: Over- and Underprediction,” a positive residual means the line underpredicted for that case, and a negative residual means the line overpredicted.

A useful new habit is to perform a residual audit before writing an interpretation. Check the subtraction order, check that the sign agrees with the comparison between \(y\) and \(\hat{y}\), check that the units match the response, and then explain the result in context. These checks reinforce one another: if the observed response is smaller than the prediction, the residual must be negative.

A Four-Part Residual Audit

This audit is not a different residual formula. It is a practical way to prevent a calculation mistake from turning into an incorrect interpretation. It also helps when software supplies the residual directly: check that the reported value’s sign agrees with the observed and predicted responses when those values are available.

The Sign Describes the Response Difference

A negative residual is not a “negative response,” and it does not mean that the prediction itself was negative. It means that the observed response was less than the predicted response. Likewise, a positive residual does not mean that the response is necessarily large in an absolute sense; it means only that the observation is above the prediction for that case.

It is often helpful to state the comparison in ordinary language before interpreting the residual. If a model predicts 8.0 hours of sleep and the person sleeps 7.7 hours, the observed value is 0.3 hour below the prediction. That comparison confirms a residual of \(-0.3\) hour. The residual’s sign and the words “below the prediction” agree.

Do not confuse the residual with the vertical distance alone. The absolute value \(|y-\hat{y}|\) gives the distance from the point to the fitted line, without direction. The signed residual gives both direction and size. In “Comparing Residuals Across Observations,” absolute residuals were useful for comparing which predictions were closer; for an interpretation that asks whether the model overpredicted or underpredicted, keep the sign.

The response units matter even when the numerical value seems familiar. If \(x\) is in hours and \(y\) is in minutes, a residual is in minutes because it is a difference between response values. It is not automatically in the predictor’s units, nor is it automatically a percentage. If converting to another unit is useful, state the conversion and keep it equivalent to the original residual.

Do not add units to a bare number without checking what the variables represent. For example, a residual of \(-1.4\) for a model predicting mass in kilograms means 1.4 kilograms below the prediction, not 1.4 predictor units below an \(x\)-value. The regression line predicts \(y\), and the residual measures the discrepancy in \(y\).

Output, Rounding, and Misleading Descriptions

A residual may be copied from computer output or calculated using displayed coefficients. If the coefficients are rounded, a hand-calculated residual can differ slightly from the software’s value, because software may use more digits internally. Before calling an output value wrong, check the data row, the data order, the equation and its precision, and the subtraction convention. This is the same careful matching emphasized in “Reading Residuals From Computer Output.”

Small rounding differences do not justify changing the sign or dropping the units. If an output reports \(0.02\) meters, report the units and interpret the positive sign; do not call it 2 centimeters unless you make the conversion clear. If the task requests an interpretation rather than a calculation, avoid reporting excessive digits that suggest more precision than the output supports.

Common Mistakes and AP Exam Tips

• Subtracting in reverse. Writing \(\hat{y}-y\) gives the opposite sign from the residual. A full-credit calculation shows \(y-\hat{y}\), observed minus predicted.
• Interpreting the sign backwards. Positive means observed above predicted and underprediction; negative means observed below predicted and overprediction. Verify this by comparing the two response values.
• Using predictor units. A residual is measured in the response variable’s units. Name the response and its units before writing the interpretation.
• Calling the residual a distance without direction. “The point is 5 minutes from the line” gives only a size. If the question asks what the residual means, retain the sign and say whether the model overpredicted or underpredicted.
• Calling a negative residual a negative observation. The response itself may be positive. The negative value describes the difference between the observed and predicted responses.
• Confusing an individual residual with \(s\). A residual describes one case; as in “Standard Deviation of the Residuals, \(s\),” \(s\) describes the typical size of prediction errors for the fitted model.

A strong AP response usually includes the case or individual, the signed residual with response units, and a sentence explaining what was observed compared with what was predicted. Avoid a bare statement such as “the residual is \(-5\).” Prefer: “For this repair, the observed time was 5 minutes less than predicted, so the model overpredicted the repair time by 5 minutes.” That sentence gives the direction, size, units, and context.