Reading a Numeric Predictor's Slope

What a Numeric Predictor’s Slope Says

In a one-predictor linear equation, the slope describes how the fitted outcome changes as the numeric predictor increases. Its interpretation is compact, but it is easy to misread: the slope is not the predicted outcome itself, and its units are not simply the outcome’s units. A careful interpretation states the direction and size of the fitted change per one unit of the predictor, using the measurement units from the situation.

Here we focus on a linear equation with one numeric predictor. The slope describes the straight-line fitted relationship represented by that equation. It does not, by itself, show that changing the predictor causes the outcome to change.

Definition: In the one-predictor fitted equation \(\hat{y}=b_0+b_1x\), \(x\) is the numeric predictor, \(\hat{y}\) is the fitted outcome, \(b_0\) is the intercept, and \(b_1\) is the slope. The slope \(b_1\) is the change in fitted outcome associated with a one-unit increase in \(x\). Its units are outcome units per predictor unit.

From the Equation to a One-Unit Change

The slope interpretation follows directly from the equation. At predictor value \(x\), the fitted outcome is \(b_0+b_1x\). At a value one unit higher, \(x+1\), it is \(b_0+b_1(x+1)\). Subtracting the first fitted value from the second gives the fitted change:

\begin{aligned} \hat{y}(x+1)-\hat{y}(x) &= [b_0+b_1(x+1)]-[b_0+b_1x] \\ &= b_0+b_1x+b_1-b_0-b_1x \\ &= b_1. \end{aligned}

The intercept and the terms involving \(x\) cancel, leaving the slope. This difference is the same for every \(x\) because the equation is linear: it has one constant slope throughout the range represented by the line. Thus, if \(b_1\) is positive, the fitted outcome increases by \(b_1\) outcome units per predictor unit. If it is negative, the fitted outcome decreases by \(|b_1|\) outcome units per predictor unit. If it is zero, the fitted line is flat.

The population version of a one-predictor linear model can be written as \(E(Y\mid X=x)=\beta_0+\beta_1x\). Here, \(\beta_1\) describes the change in the model’s mean outcome for a one-unit increase in the predictor. In a fitted equation, \(b_1\) is the estimated slope, so its interpretation refers to the change in the fitted outcome. The arithmetic interpretation of a fitted slope does not require a significance test. Assessing uncertainty or testing a slope requires additional methods.

Worked Example: Study Time and Fitted Quiz Score

Suppose a hypothetical class is summarized by the fitted equation \(\hat{y}=72+3.6x\), where \(x\) is study time in hours and \(\hat{y}\) is fitted quiz score in points. Interpret the slope.

The slope is \(b_1=3.6\). Its units are quiz-score points per study hour. Therefore, for each additional hour of study, the equation’s fitted quiz score increases by 3.6 points, on average in the fitted linear relationship. The statement describes fitted values, not a promise that every student who studies one extra hour will gain exactly 3.6 points.

The one-unit comparison verifies the interpretation. For a change from 4 to 5 study hours, the fitted change is

\begin{aligned} [72+3.6(5)]-[72+3.6(4)] &=90-86.4 \\ &=3.6\text{ points}. \end{aligned}

The calculation uses a one-hour increase and gives a 3.6-point increase in the fitted score. It illustrates the slope; it does not establish that study time caused a score change.

Direction and Units Matter

A slope’s sign gives the direction of the fitted linear relationship. A positive slope means larger predictor values correspond to larger fitted outcomes. A negative slope means larger predictor values correspond to smaller fitted outcomes. The sign alone does not tell whether the change is large in practical terms; that depends on the slope’s magnitude and the units used for both variables.

Always name both units. If the outcome is measured in minutes and the predictor in kilometres, a slope of 0.5 means 0.5 minutes per kilometre. It does not mean 0.5 kilometres per minute. Reversing the order of the units would describe a different quantity.

Worked Example: Exercise Time and Fitted Resting Pulse

Consider a hypothetical fitted equation \(\hat{y}=74.2-1.25x\), where \(x\) is weekly exercise time in hours and \(\hat{y}\) is fitted resting pulse rate in beats per minute. State and check the slope interpretation.

The slope is \(-1.25\) beats per minute per exercise hour. Its negative sign means that each one-hour increase in weekly exercise time corresponds to a 1.25-beat-per-minute decrease in the fitted resting pulse rate.

For example, compare 6 and 7 exercise hours per week:

\begin{aligned} [74.2-1.25(7)]-[74.2-1.25(6)] &=65.45-66.70 \\ &=-1.25\text{ beats per minute}. \end{aligned}

The fitted value is 1.25 beats per minute lower at 7 hours than at 6 hours. This is a description of the fitted association. The equation alone does not show that additional exercise causes a lower resting pulse; other variables or the way the data were collected may matter.

The Predictor’s Unit Defines “One Unit”

A one-unit increase means one unit in the way the predictor is recorded in the equation. If distance is recorded in kilometres, one unit is one kilometre. If it is recorded in tens of kilometres, one unit is ten kilometres. The coefficient’s numerical value changes when the predictor’s scale changes, even though the underlying fitted line can describe the same relationship.

For instance, a slope of 5 minutes per 10 kilometres is equivalent to 0.5 minutes per kilometre. The interpretations differ in the size of the predictor step, not in the relationship they describe. Read the variable’s measurement scale before attaching a verbal meaning to the number.

Worked Example: Reading a Slope When Distance Is Scaled

Suppose a hypothetical travel-time equation is \(\hat{y}=12+5.4x\), where \(\hat{y}\) is fitted travel time in minutes and \(x\) is distance measured in 10-kilometre units. Interpret the slope in the recorded units, then express it per kilometre.

The slope is 5.4 minutes per unit of \(x\), and each unit of \(x\) represents 10 kilometres. Thus, each additional 10 kilometres corresponds to a 5.4-minute increase in fitted travel time. For a one-unit increase, such as a change from 2 to 3 units, the fitted difference is

[12+5.4(3)]-[12+5.4(2)] =28.2-22.8 =5.4\text{ minutes}.

Since 10 kilometres correspond to 5.4 minutes, the equivalent fitted change per kilometre is

\frac{5.4\text{ minutes}}{10\text{ kilometres}} =0.54\text{ minutes per kilometre}.

Both descriptions refer to the same fitted relationship. Saying “5.4 minutes per kilometre” would be incorrect because \(x\) is measured in 10-kilometre units, not one-kilometre units.

What the Slope Does Not Say

The slope describes the fitted linear relationship between the predictor and outcome. It does not say that every observed outcome changes by exactly the slope when the predictor differs by one unit. Individual observations can lie above or below the fitted line. Nor does a slope, on its own, establish causation. A causal claim needs support from the study design and relevant assumptions, not merely a regression equation.

The linear equation also represents a constant fitted change per predictor unit over the range where the line is used. The slope should not automatically be treated as a reliable description far outside the predictor values represented in the data. A straight-line equation can produce a numerical value there without making that value a sensible or well-supported interpretation.

Key takeaway: To interpret a numeric predictor’s slope, state the fitted outcome’s change for a one-unit increase in the predictor, name both measurement units, and preserve the sign. Describe a fitted association rather than a guaranteed individual change or a causal effect.

Common Mistakes

Leaving out “per unit.” Saying “the slope is 3.6 points” is incomplete. Say “the fitted score increases by 3.6 points per additional study hour.”
Reversing the units. A slope measured in minutes per kilometre is not measured in kilometres per minute. State the outcome unit first and the predictor unit second.
Ignoring a negative sign. A slope of \(-1.25\) means the fitted outcome decreases by 1.25 units for each one-unit predictor increase, not that it increases by 1.25.
Calling the slope a causal effect. “An additional hour causes a 1.25-beat decrease” goes beyond what a fitted equation alone supports. Describe the fitted association unless the study design justifies a causal interpretation.
Using the wrong predictor scale. When the predictor is recorded in tens of kilometres, one unit represents 10 kilometres. Check the variable’s units before describing the change.

Check Your Understanding

Answer each question using the slope’s sign and measurement units.

A fitted equation relates weekly reading time \(x\), in hours, to a fitted vocabulary score \(\hat{y}\), in points: \(\hat{y}=40+2.1x\). Interpret the slope in context.
A fitted slope is \(-0.8\) kilograms per centimetre for an outcome measured in kilograms and a predictor measured in centimetres. What does a one-unit increase in the predictor mean for the fitted outcome?
A distance predictor is recorded in 5-kilometre units, and the fitted slope is 3 minutes per predictor unit. How many minutes per kilometre is this, and what does the original slope mean?
Why is it incorrect to say that every individual’s outcome must change by exactly the slope when the predictor increases by one unit?
What additional evidence, beyond a fitted slope, is needed before describing a predictor-outcome relationship as causal?