What a Regression Slope Says
A fitted regression line summarizes a linear pattern between an explanatory variable and a response variable. Its slope describes the direction and rate of change in the line: how much the predicted response changes for a one-unit increase in the explanatory variable. That meaning is useful only when stated with the variables and their units attached. A number such as 2.4, by itself, does not say what changes or how it is measured.
The explanatory variable is denoted by \(x\), and the response variable by \(y\). A fitted least-squares regression line can be written as \(\hat{y}=b_0+b_1x\), where \(\hat{y}\) is the response value predicted by the line, \(b_0\) is the intercept, and \(b_1\) is the fitted slope. This tutorial focuses on the interpretation of \(b_1\), not on how the line is calculated.
The unit of slope is the response variable’s unit divided by the explanatory variable’s unit. For example, if response values are test-score points and explanatory values are study hours, the slope is measured in points per hour. If both variables are recorded in the same unit, the slope still describes response units per explanatory-variable unit; those units should not be left implicit.
Why the Slope Is a One-Unit Change
The definition follows directly from the equation of the fitted line. At a value \(x\), the line predicts \(b_0+b_1x\). At a value one unit higher, \(x+1\), it predicts \(b_0+b_1(x+1)\). Subtracting the first prediction from the second gives the change in predicted response:
Thus, a positive slope means the line predicts a larger response at a larger explanatory-variable value; a negative slope means it predicts a smaller response. The absolute size of the slope gives the number of response units in that one-unit change. This is a property of the fitted line, not a guarantee that every observed individual follows that exact change.
Worked Example: Study Time and Quiz Score
Suppose, in a hypothetical sample of students, study time is measured in hours and quiz score is measured in points. A fitted line is \(\hat{y}=51+2.4x\), where \(x\) is study time in hours and \(\hat{y}\) is the predicted quiz score in points. Interpret its slope.
The slope is \(b_1=2.4\). Its units are quiz-score points per study hour. For a one-hour increase in study time, the line’s predicted score change is \(2.4 \text{ points/hour}\times 1 \text{ hour}=2.4 \text{ points}\).
In context: For each additional hour of study time, the fitted line predicts a quiz score 2.4 points higher, on average. The word “predicts” identifies this as the change in the fitted line’s response, rather than an assertion that every student’s score rises by exactly 2.4 points for each extra hour.
This interpretation describes a fitted association in the hypothetical data. It does not establish that studying an additional hour causes a student’s score to rise by 2.4 points. Other factors, such as prior preparation, may be related to both study time and score.
Reading Positive and Negative Slopes
The sign and size of the slope answer different parts of the interpretation. The sign gives the direction of the fitted relationship. The size gives the predicted response change for a one-unit increase in \(x\). A negative sign belongs in the mathematical slope, but it is usually clearer in a sentence to say “decreases” and report the positive magnitude of that decrease.
| Fitted slope | Direction on the fitted line | Interpretation pattern |
|---|---|---|
| \(b_1>0\) | Predicted response increases as \(x\) increases | For each additional one unit of \(x\), predicted \(y\) increases by \(b_1\) response units. |
| \(b_1<0\) | Predicted response decreases as \(x\) increases | For each additional one unit of \(x\), predicted \(y\) decreases by \(|b_1|\) response units. |
| \(b_1=0\) | The fitted line is horizontal | A one-unit increase in \(x\) corresponds to no change in the line’s predicted \(y\). |
A slope is not the predicted response itself. It is also not automatically a percentage or a proportion. Its interpretation depends on what one unit means for \(x\), what is measured by \(y\), and the units in which both variables were recorded. The next examples keep those details visible.
Worked Example: Outdoor Temperature and Heating Energy
Suppose a hypothetical set of daily records is summarized by a fitted line relating outdoor temperature \(x\), in degrees Celsius, to a building’s heating energy use \(\hat{y}\), in kilowatt-hours. The fitted slope is \(-1.8\) kilowatt-hours per degree Celsius. Interpret the slope.
A one-unit increase in \(x\) is an increase of \(1^\circ\text{C}\). The predicted response change is \((-1.8 \text{ kWh}/^\circ\text{C})\times(1^\circ\text{C})=-1.8 \text{ kWh}\).
In context: For each \(1^\circ\text{C}\) increase in outdoor temperature, the fitted line predicts 1.8 fewer kilowatt-hours of heating energy use per day. The negative slope indicates a decrease in the response as temperature increases. The units reduce to kilowatt-hours because the degree-Celsius unit in the slope is multiplied by a one-degree-Celsius change.
This is a description of the fitted line for the hypothetical records, not proof that temperature alone accounts for daily energy use. A slope interpretation should not turn a pattern in data into an unsupported causal conclusion.
Keep the Interpretation Within Its Meaning
Slope describes the fitted linear trend over the data used to fit the line. It does not say that the line predicts every observed response exactly: individual observations can lie above or below it. Nor does the slope alone tell how well the line predicts responses. A steep fitted slope can coexist with considerable scatter around the line, and a shallow slope can coexist with relatively little scatter.
A slope interpretation should also be tied to the actual variables, rather than replacing them with vague phrases such as “\(y\) goes up.” Name the response and explanatory variables, state which one increases, give the predicted change in the response, and include units. If the observations cover only a restricted range of \(x\), avoid implying that the same linear pattern must continue far beyond that range. The slope is the line’s rate of change; whether the line is useful outside the data’s range is a separate question.
Worked Example: Age and Length of a Hypothetical Hospital Stay
Suppose a hypothetical observational data set records patient age in years and hospital stay in days. A fitted line has a slope of \(0.06\) days per year. Interpret the slope carefully.
The explanatory variable is age, and the response variable is length of stay. One unit of age is one year, so the predicted response change for a one-year increase is \((0.06 \text{ days/year})\times(1 \text{ year})=0.06 \text{ days}\).
In context: For each additional year of age, the fitted line predicts a hospital stay 0.06 days longer. This is the predicted change in the line, not a claim that a particular patient’s stay will be 0.06 days longer than another patient’s stay.
The data are observational, so this slope does not show that age causes a longer stay. Differences in health conditions or other characteristics could contribute to the pattern. The interpretation states the fitted association and its units without assigning a cause.
A Reliable Interpretation Process
A short sequence of checks helps turn a numerical coefficient into a complete sentence. Start by identifying which variable is explanatory and which is the response. Then check the slope’s sign and units before describing its change. Finally, make clear that the statement concerns the fitted prediction and is not, by itself, a causal conclusion.
Identify what \(x\) measures and what response \(y\) measures. Do not swap their roles.
A positive slope indicates an increase in fitted response; a negative slope indicates a decrease. Express the units as response units per explanatory-variable unit.
Use “for each additional one [unit] of [explanatory variable], the fitted line predicts [change] [response units] in [response variable].”
Describe a fitted trend, not an exact change for every observation or a causal effect that the data do not establish.
Common Mistakes
- Leaving out the units. Saying “the slope is 2.4” gives no contextual meaning. State which response units change per one explanatory-variable unit.
- Reversing the variables. The slope is response change per explanatory-variable unit, not explanatory-variable change per response unit.
- Dropping the direction. If the slope is negative, describe a predicted decrease. Reporting only its absolute value can reverse the meaning.
- Claiming an exact individual change. The slope describes the fitted line’s predicted change, not necessarily what happens to each person or object.
- Claiming causation from a fitted association. A regression slope alone does not show that changing \(x\) causes a change in \(y\).
- Calling the slope a percent without justification. A slope of 2.4 means 2.4 response units per explanatory-variable unit, not a 2.4% increase unless the variables and model give that specific interpretation.
Check Your Understanding
For each question, focus on the meaning of the fitted slope and keep the variables and units explicit.
- A fitted line relates weekly exercise minutes to resting heart rate in beats per minute and has slope \(-0.03\) beats per minute per exercise minute. Write an interpretation in context.
- A model predicts a crop yield in kilograms from rainfall in millimeters, with slope \(0.8\) kilograms per millimeter. What does the slope say about a one-millimeter increase in rainfall?
- Why is “each person’s response decreases by 1.8” not an adequate interpretation of a slope of \(-1.8\)?
- A fitted slope is stated as 5 score points per study hour. Which variable is the response, and which is the explanatory variable?
- Why does an observational regression slope not by itself establish that changing the explanatory variable causes the predicted response change?