Make Every Regression Sentence Specific
A regression sentence can be numerically correct and still be incomplete. Saying “the slope is 2.4” does not tell a reader which variables are involved, what the units are, or what the number means. In “Variables, Units, and Meaning in a Regression Model,” we identified the individuals, explanatory variable, and response variable. Here, we use those details every time we describe a slope, intercept, or prediction.
A fitted regression line is often written as \(\hat{y}=a+bx\). The explanatory variable is \(x\), the response variable is \(y\), \(a\) is the intercept, and \(b\) is the slope. The fitted value \(\hat{y}\) is the response predicted by the line at a particular value of \(x\). The variables and units give each part of the equation its meaning.
A dependable interpretation has four parts: who or what the cases are, which variable is explanatory, which variable is the response, and what units the number uses. For example, if each case is a delivery, \(x\) is delivery distance in kilometers, and \(y\) is delivery time in minutes, the slope’s units are minutes per kilometer. The predicted time itself is measured in minutes.
Sentence Frames for Slope, Intercept, and Prediction
Use a sentence frame as a checklist, not as a substitute for understanding the situation. Fill in the names and units rather than leaving the answer at “\(y\) goes up by \(b\).”
- Slope: For each additional one [explanatory-variable unit], the model predicts that [response variable] changes by \(b\) [response-variable units], on average, for the cases represented by the data.
- Intercept: When [explanatory variable] is 0 [explanatory-variable units], the model predicts [response variable] to be \(a\) [response-variable units]. Then consider whether zero is relevant to the setting and supported by the data’s range.
- Prediction: At [a stated value and units of the explanatory variable], the model predicts [response variable] to be \(\hat{y}\) [response-variable units] for [the relevant kind of case].
The slope is a predicted change in the response for a one-unit increase in the explanatory variable. If \(b\) is negative, use “decreases” or “is lower by” rather than “increases.” The slope describes the fitted linear relationship; in an observational study, it does not by itself say that changing \(x\) causes \(y\) to change.
The intercept is the fitted value when \(x=0\), because substituting zero into \(\hat{y}=a+bx\) gives \(\hat{y}=a\). That makes the algebraic interpretation straightforward, but it does not guarantee that the interpretation is practically useful. Zero might be impossible, outside the observed range, or unrelated to the question being studied. As in “Using a Model to Make Decisions,” be alert to when a calculation extends beyond the values represented in the data.
A prediction is a fitted response for a specified explanatory-variable value. Name that value and its units, then state the predicted response with its units. Say “the model predicts,” not “the response will be.” A fitted value is not an observed outcome or a guarantee for an individual case.
Worked Examples: From Equation to Context
Worked Example: Predicting Delivery Time
A fictional courier service records individual deliveries. For each delivery, \(x\) is the route distance in kilometers and \(y\) is the delivery time in minutes. The fitted line is \(\hat{y}=7+2.4x\). The distances in the data ranged from 2 to 15 kilometers.
Interpret the slope. The slope is \(2.4\). Its units are minutes per kilometer: the response is measured in minutes and the explanatory variable in kilometers. A complete interpretation is: For deliveries represented by these data, each additional kilometer of route distance is associated with a predicted increase of 2.4 minutes in delivery time, on average. This describes the fitted association, not proof that distance alone causes the difference.
Interpret the intercept. The intercept is 7 minutes. At a route distance of 0 kilometers, the line predicts a delivery time of 7 minutes. However, the recorded distances begin at 2 kilometers, so 0 is outside the observed range. The algebraic meaning is clear, but the data do not establish that this prediction is dependable for a zero-kilometer route. Avoid presenting it as a meaningful measured baseline without more context.
Make a prediction. For a delivery with a route distance of 8 kilometers, substitute \(x=8\):
A complete prediction statement is: For a delivery on an 8-kilometer route, the model predicts a delivery time of 26.2 minutes. Eight kilometers is within the observed range of 2 to 15 kilometers. The prediction is still an estimate for a delivery, not a promise that every such delivery takes exactly 26.2 minutes.
Worked Example: Describing Seedling Growth
A fictional garden program measures seedlings at different times after transplanting. Each case is one seedling. The explanatory variable \(x\) is time since transplanting in weeks, and the response variable \(y\) is seedling height in centimeters. The fitted line is \(\hat{y}=12+1.8x\). Measurements cover weeks 0 through 8.
Interpret the slope. The slope is \(1.8\) centimeters per week. A contextual statement is: For seedlings represented by these data, each additional week after transplanting is associated with a predicted increase of 1.8 centimeters in height, on average. The phrase “per week” is essential: 1.8 is not a height, but a predicted height change for each additional week.
A reader might also ask about a three-week difference. The line predicts a difference of \(1.8\) centimeters per week for each of three weeks:
Thus, for two times three weeks apart within the modeled range, the fitted line predicts a height difference of 5.4 centimeters. This is a comparison of predicted heights, not a claim that every pair of seedlings differs by exactly that amount.
Interpret the intercept. At week 0, the line predicts a height of 12 centimeters. Week 0 is the time of transplanting and is included in the data range, so the intercept has a direct interpretation in this setting: At the time of transplanting, the model predicts a height of 12 centimeters for the seedlings represented by the data. It remains a model prediction, rather than a claim that every seedling was exactly 12 centimeters tall.
Make a prediction. At week 5, the fitted height is:
A complete statement is: For a seedling represented by these data, the model predicts a height of 21 centimeters five weeks after transplanting. The predictor value is inside the observed range, and the response is stated in centimeters. The line summarizes the pattern across seedlings; it does not guarantee the height of a particular seedling.
Worked Example: Interpreting a Fundraising Model
A fictional school club examines past fundraising campaigns. Each case is one campaign. The explanatory variable \(x\) is the number of social media posts during the campaign, and the response variable \(y\) is the money raised in dollars. The fitted line is \(\hat{y}=45+18x\). The campaigns in the data had between 2 and 12 posts.
Interpret the slope. The slope is 18 dollars per post. A clear statement is: For campaigns represented by these data, each additional social media post is associated with a predicted increase of 18 dollars in money raised, on average. This wording identifies both variables and makes clear that the dollar amount is a predicted change, not a guaranteed return from publishing one more post.
Interpret the intercept. The intercept is 45 dollars, so the line predicts 45 dollars raised when the number of posts is 0. But the campaigns observed had at least 2 posts. The model’s intercept is therefore outside the observed predictor range. You can state what the intercept means mathematically, but should qualify its practical relevance: the data do not show whether the fitted relationship is dependable for a campaign with no posts.
Make a prediction. For a campaign with 7 posts, the model predicts:
The contextual statement is: For a campaign represented by these data with 7 social media posts, the model predicts that 171 dollars will be raised. Seven posts is within the observed range of 2 to 12, and the predicted response is expressed in dollars. The word “predicts” matters: the actual amount raised could differ.
This is an observational relationship in past campaigns. The slope alone does not establish that adding posts causes fundraising to increase by 18 dollars per post. Other campaign features could be related to both the number of posts and the money raised.
Check the Units and the Scope of the Sentence
A quick unit check catches many incomplete interpretations. In the delivery example, the response is minutes and the explanatory variable is kilometers, so the slope must be minutes per kilometer. The intercept and predicted delivery times are minutes. In the fundraising example, the slope is dollars per post, while the intercept and predicted amount are dollars.
Before submitting a regression interpretation, ask:
- Did I name the variables? Replace bare \(x\) and \(y\) with the real explanatory and response variables.
- Did I state the units correctly? A slope uses response units per explanatory-variable unit; a predicted response uses response units.
- Did I describe a prediction rather than a certainty? Use “the model predicts,” and identify the case or cases the prediction concerns.
- Is zero meaningful for the intercept? Explain the fitted value at \(x=0\), then consider whether that value fits the setting and the observed range.
- Am I implying causation? For an observational relationship, use associational language such as “is associated with a predicted increase,” not “causes an increase.”
Keep the interpretation at the level of the data. A model fitted to group averages predicts group averages, as explained in “Group Averages Versus Individual Predictions.” If the cases are campaigns, days, or seedlings, do not quietly turn a group-level statement into a claim about a different kind of individual.
Common Mistakes and AP Exam Tips
Several small wording choices make the difference between a vague answer and a complete one.
- Reporting only a coefficient. “The slope is 2.4” gives no contextual meaning. State the predicted response change for a one-unit increase in the named explanatory variable and include both units.
- Reversing the slope’s units. If \(y\) is in minutes and \(x\) is in kilometers, the slope is minutes per kilometer—not kilometers per minute. Check which variable is the response.
- Giving the slope as the response value. A slope of 1.8 centimeters per week is not a predicted height of 1.8 centimeters. It describes predicted height change per week.
- Calling the intercept the starting value without checking the setting. The intercept is the model’s predicted response at \(x=0\). If zero is impossible, unobserved, or outside the data’s range, say so instead of treating the value as a dependable baseline.
- Leaving units off a prediction. “The prediction is 26.2” is incomplete. Name the response and say 26.2 minutes, dollars, centimeters, or the appropriate response units.
- Claiming certainty or causation. “Every delivery will take 26.2 minutes” and “one more post causes 18 more dollars” overstate what a fitted line establishes. Say what the model predicts and describe an observational slope as an association.
- Ignoring the cases. State whether the prediction concerns a delivery, a seedling, a campaign, or another case. The fitted line predicts the response for cases like those represented by its data.
A full-credit sentence typically identifies the cases, names the explanatory and response variables, gives the numerical meaning with units, and qualifies the claim appropriately. For an intercept or prediction outside the observed range, add a clear caution about interpreting the result. These details make the statistical meaning understandable without requiring the reader to reconstruct it from the equation.
Check Your Understanding
For each question, write a contextual sentence and include the relevant units or limitation.
- A model predicts quiz score \(y\), in points, from tutoring sessions \(x\): \(\hat{y}=52+4x\). Interpret the slope in context.
- For the model in question 1, what does the intercept mean? If the data included students with zero sessions, explain why that helps assess the intercept’s relevance.
- A fitted line predicts daily water use in liters from the number of people in a household. What units should the slope have, and what units should a predicted water-use value have?
- A model predicts that a 6-kilometer route takes 19 minutes. Write a complete prediction statement that makes clear the value is predicted rather than guaranteed.
- A regression model for fundraising campaigns predicts more money for campaigns with more social media posts. Why is “each extra post causes the predicted increase” too strong based on the fitted line alone?