A Slope Describes a Fitted Pattern, Not Automatically a Cause
In “Interpreting Slope in Context,” the slope \(b\) was described as the change in the predicted response for a one-unit increase in the predictor. That interpretation describes the regression line. It does not, by itself, tell us what would happen if someone deliberately changed the predictor.
This distinction matters especially when data are observational. In an observational study, researchers record or measure the explanatory variable and response without assigning the explanatory variable to cases. A regression line can summarize an association in those data, but an association alone does not establish that changing the predictor causes the response to change. As discussed in “Why Correlation Does Not Imply Causation,” a strong pattern is not proof of a causal relationship.
The problem is not that slope interpretations use words such as “increase” or “decrease.” Those words can accurately describe the direction of the fitted prediction. The problem is wording that makes the predictor sound as though it produces the response change. Compare these two statements:
- Model-based: “For each additional hour of the predictor, the regression line predicts a lower response by a stated amount.”
- Causal: “Adding an hour of the predictor lowers the response by a stated amount.”
The first describes what the fitted line predicts. The second sounds like an intervention: deliberately adding an hour would make the response fall. An observational regression alone does not justify that conclusion.
A Safe Pattern for an Observational Slope Statement
When the data are observational, keep the regression line at the center of the interpretation. Name the cases, predictor, response, units, and direction. Make clear that the amount is a change in the predicted response, not a guaranteed change in every case or evidence of a cause.
For a positive slope, the predicted response increases as the predictor increases. For a negative slope, the predicted response decreases. The slope’s units remain response units per predictor unit, as explained in “Slope Units and Rates of Change.” But adding “the regression line predicts” is especially helpful when the sentence might otherwise sound causal.
The phrase “among the cases in the data” also keeps the claim appropriately scoped. A sample regression line describes the fitted pattern in the observed data. Be cautious about broadening the sentence to all people, all places, or all future cases unless the study design supports that generalization.
A causal explanation might be plausible, but other explanations can also contribute to an observed association. In “Lurking Variables and Confounding,” a lurking variable was defined as a variable not included in the analysis that may affect the relationship between the predictor and response. Mentioning a plausible lurking variable can explain why an observational slope does not settle the causal question.
Worked Examples
Worked Example: Screen Time and Sleep in a Survey
A fictional survey records daily recreational screen time and hours of sleep per night for a group of teenagers. Let \(x\) be screen time in hours per day and \(y\) be sleep in hours per night. The fitted regression line is:
A student writes, “Every additional hour of screen time causes teenagers to get 0.22 fewer hours of sleep.” Revise the statement.
The slope is \(-0.22\), measured in hours of predicted sleep per additional hour of screen time. Its negative sign says the fitted sleep prediction goes down as screen time increases. A careful interpretation is: Among the teenagers in this survey, for each additional hour of daily recreational screen time, the regression line predicts 0.22 fewer hours of sleep per night.
The original sentence overstates what the data establish. The survey recorded screen time and sleep; it did not assign teenagers different amounts of screen time. The fitted association does not show that screen time caused the lower sleep prediction. For example, school schedules or other routines might be related to both screen time and sleep. That possibility does not prove a particular alternative explanation; it reminds us that the observed relationship alone does not establish causation.
The slope is about the line’s predicted response, not a guarantee that every teenager who uses a screen one hour longer will sleep 0.22 fewer hours. Individual responses can differ from their predicted values, as covered in “Predicted Change Versus Actual Change.” A full interpretation therefore says “the regression line predicts” rather than promising an individual outcome.
Worked Example: Tutoring Hours and Test Scores
A fictional school records the number of optional tutoring hours students attend and their scores on a later assessment. Students choose whether to attend tutoring; the school does not assign them to different tutoring amounts. The fitted line is:
Here, score is measured in points and tutoring time in hours. A report says, “Each hour of tutoring raises a student’s score by 2.4 points.” What is a more appropriate interpretation?
The slope is \(2.4\) points per tutoring hour. In context: Among the students whose records were analyzed, for each additional hour of tutoring attended, the regression line predicts a score that is 2.4 points higher. This describes a positive linear association between tutoring hours and predicted assessment score in the observed data.
“Raises” implies tutoring itself produces the increase. But students selected their own tutoring hours, so the data are observational. Students who attend more tutoring might also differ in prior achievement, motivation, available study time, or other factors related to scores. Prior achievement, for example, could be a lurking variable if it is related both to tutoring attendance and to the later score. This is the kind of possible alternative discussed in “Lurking Variables and Confounding.”
The slope does not say that every student who attends one more hour will gain exactly 2.4 points. It also does not establish what would happen if otherwise comparable students were assigned different tutoring amounts. The model predicts a higher score for a one-unit difference in tutoring hours along its fitted line; that is not the same as demonstrating a causal gain.
Worked Example: Sunlight and Plant Growth
In a fictional observational project, a gardener measures the daily hours of direct sunlight and weekly growth for plants already growing in several garden plots. The fitted line predicts weekly growth in centimeters:
A student summarizes the slope as, “An extra hour of sunlight makes each plant grow 0.7 centimeters more per week.” Rewrite and explain the statement.
The slope \(0.7\) has units of centimeters of predicted growth per additional hour of sunlight. A suitable interpretation is: Among the plants in these observed plots, for each additional hour of daily direct sunlight, the regression line predicts 0.7 centimeters more weekly growth. The positive slope describes the direction of the fitted association.
The phrase “makes each plant grow” is not supported by this observational slope. The project measured plants in plots with different sunlight; it did not randomly assign sunlight conditions. Water availability, soil differences, or plant variety could be related to both sunlight and growth. These are possible explanations to consider, not established facts about this fictional project.
Notice that removing causal language does not require hiding the slope’s direction. “Predicts 0.7 centimeters more” clearly states the positive fitted change, while keeping the claim about the regression line. It does not claim that every plant follows the prediction or that changing sunlight would produce that amount of additional growth.
Choose Words That Match the Study Design
Before writing a slope interpretation, ask how the predictor values arose. If researchers observed or recorded the predictor without assigning it, use language of association and prediction. If the study randomly assigns treatments, a well-conducted randomized experiment may support a causal conclusion. Even then, the causal interpretation comes from the design and evidence—not from the slope alone.
For this tutorial’s observational situations, avoid verbs that assert a cause-and-effect relationship. Words such as “causes,” “makes,” “produces,” “improves,” “reduces,” “raises,” and “leads to” can turn a description of the fitted pattern into an unsupported causal claim. Prefer “is associated with,” “the line predicts,” or “the fitted response is higher/lower for cases with a larger predictor value.”
Some verbs are not automatically wrong in every sentence. “Increases” is suitable in “the predicted response increases as \(x\) increases,” because the sentence explicitly describes the line. But “increasing \(x\) increases \(y\)” may sound like a causal instruction. Adding “the regression line predicts” makes the intended meaning much clearer.
| Wording to avoid for observational data | Safer slope wording |
|---|---|
| “More screen time causes less sleep.” | “The line predicts less sleep for cases with more screen time.” |
| “Tutoring raises scores.” | “Tutoring hours and predicted scores have a positive association in these data.” |
| “Sunlight makes plants grow faster.” | “The fitted line predicts greater weekly growth for plots with more sunlight.” |
A safer wording is not a claim that the predictor definitely has no causal role. It is a statement that the observational regression, by itself, does not establish that role. This distinction is important: “causation has not been demonstrated” does not mean “causation is impossible.”
Common Mistakes and AP Exam Tips
- Turning a predicted change into a causal effect. “One more hour makes the response rise” implies an intervention. For observational data, say “for each additional hour, the regression line predicts a response that is higher by…”
- Leaving out the response or predictor units. A slope is not just a number. State response units per predictor unit, such as points per tutoring hour or hours of sleep per screen-time hour.
- Making an individual guarantee. The slope describes how the fitted prediction changes, not what every individual will experience. Include “the regression line predicts” and do not promise the same actual change for each case.
- Claiming the association proves there is no causal relationship. An observational study does not establish causation on its own, but that is not proof that the predictor can have no causal role. Limit the conclusion to what the design supports.
- Ignoring possible lurking variables. Do not invent a definite explanation from a variable that was not studied. Instead, name a plausible variable and explain how it could relate to both the predictor and response.
- Making the conclusion broader than the data. “Teenagers” or “all plants” may be too broad if the line was fitted to a particular sample or set of plots. Identify the cases represented by the data.
For full credit, an AP-style slope interpretation should name the predictor and response, include the direction and units, and say what the line predicts for a one-unit increase in the predictor. When the data are observational, avoid causal verbs and, if relevant, state that the association does not establish causation. A concise, careful sentence is stronger than a dramatic claim the study design cannot support.
Check Your Understanding
For each question, focus on what the observational regression supports—not on what might happen under an intervention.
- A survey-based line predicts daily exercise minutes from hours of sleep, with slope \(5.2\) minutes per sleep hour. Write a careful slope interpretation without implying causation.
- A student says, “Each extra kilometer of commute makes travel cost rise by $1.25,” based on observational commute and cost data. Identify the causal wording and revise the statement.
- Why is “the regression line predicts a lower response” more appropriate than “the predictor reduces the response” for an observational study?
- Name one plausible lurking variable for an observed association between tutoring hours and test scores. Explain how it could be related to both variables.
- Does saying “the observational data do not establish causation” mean the predictor cannot cause the response? Explain the distinction.