Slope Units and Rates of Change

A Slope Is a Rate With Units

In “Interpreting the Slope in Context,” you learned that the slope $b$ in $\hat{y}=a+bx$ describes how the line’s predicted response changes for a one-unit increase in the predictor. Here we focus on what the slope’s units tell you about that rate of change.

The slope’s units come from the units of the two variables. The response $y$ is measured in the units on the top of the ratio, and the predictor $x$ is measured in the units on the bottom. Thus, if $y$ is measured in dollars and $x$ in square feet, the slope is measured in dollars per square foot. “Per” means “for each one”: the slope describes the change in predicted dollars for each additional square foot.

Definition: The units of a regression slope are response units per predictor unit. They express the rate of change in the line’s predicted response for a one-unit increase in the predictor.

You can check the units by thinking of a change in predicted response divided by a change in predictor:

\text{slope units}= \frac{\text{response units}}{\text{predictor units}}

This is a units check, not a claim that the slope is generally the response value divided by the predictor value. The slope describes change in predicted response per change in predictor. The response-to-predictor ratio $y/x$ usually is not the slope.

A slope can therefore have units such as dollars per square foot, kilometers per hour, or kilowatt-hours per degree Celsius. The words after “per” matter: they name the predictor unit that corresponds to the change described by the slope. When interpreting the slope, identify both variables and units, and say whether the predicted response increases or decreases.

Reading the Units From a Regression Equation

In the equation $\hat{y}=a+bx$, the predicted response $\hat{y}$ has the same units as the response variable $y$. The term $bx$ must also have those response units so that it can be added to $a$. As a result, the slope $b$ must have response units divided by predictor units.

For instance, if $x$ is area in square feet and $y$ is sale price in dollars, multiplying a slope in dollars per square foot by an area in square feet gives dollars. The square-foot units cancel, leaving the units of the predicted price. This unit check works whether the slope is positive or negative.

Formula: If $x$ has units $U_x$ and $y$ has units $U_y$, then $b$ has units $U_y$ per $U_x$. In an interpretation, state the response change and predictor increase using these units.

Remember that the line’s predicted response changes at this rate; individual observations do not have to change by exactly the same amount. As discussed in “Predicted Change Versus Actual Change,” actual responses can differ from their predicted values. A slope describes the fitted line, not a guarantee about every individual case.

Worked Examples: Slope Units as Rates

Worked Example: Home Size and Sale Price

A fictional housing analysis uses area $x$, measured in square feet, to predict sale price $\hat{y}$, measured in dollars. The fitted line is:

\hat{y}=42{,}000+185x

Identify the slope units and interpret the slope in context. Then verify its rate using two area values.

The response is measured in dollars and the predictor in square feet, so the slope units are dollars per square foot:

\frac{\text{dollars}}{\text{square foot}}

To check the rate, calculate the predicted sale prices for homes of 1,000 and 1,200 square feet:

\hat{y}(1{,}000)=42{,}000+185(1{,}000)=227{,}000\text{ dollars}

\hat{y}(1{,}200)=42{,}000+185(1{,}200)=264{,}000\text{ dollars}

The predictor increased by $1{,}200-1{,}000=200$ square feet, and the predicted response increased by $264{,}000-227{,}000=37{,}000$ dollars. The rate is:

\frac{37{,}000\text{ dollars}}{200\text{ square feet}} =185\text{ dollars per square foot}

This agrees with the slope in the equation. In context: For each additional square foot of area, the line predicts a sale price that is $185 higher, on average, for homes represented by this model. Here “on average” describes the predicted response on the line; it does not mean every additional square foot changes an individual home’s actual sale price by exactly $185.

Notice that the slope is not the predicted price divided by the home’s area. For example, the predicted price of a 1,000-square-foot home divided by 1,000 square feet would be $227 per square foot. That is an average price-to-area ratio at that point, not the line’s slope of $185 per additional square foot.

Worked Example: Courier Distance and Time

A fictional delivery company models the distance $d$, in kilometers, traveled by a courier over time $t$, in hours. The fitted line is:

\hat{d}=68t

What are the slope units, and what rate does the slope describe? Also express the slope in meters per minute.

Distance is the response and is measured in kilometers; time is the predictor and is measured in hours. Therefore, the slope units are kilometers per hour. To verify the rate, compare predictions at $t=1.5$ hours and $t=4$ hours:

\hat{d}(1.5)=68(1.5)=102\text{ kilometers}

\hat{d}(4)=68(4)=272\text{ kilometers}

The predicted distance increases by $272-102=170$ kilometers during an increase of $4-1.5=2.5$ hours. So the rate is:

\frac{170\text{ kilometers}}{2.5\text{ hours}} =68\text{ kilometers per hour}

In context: For each additional hour, the line predicts that the courier’s distance traveled increases by 68 kilometers. This is a rate for the fitted line. It does not say that the courier travels exactly 68 kilometers during every individual hour.

To convert the rate to meters per minute, convert both units in the ratio. One kilometer is 1,000 meters, and one hour is 60 minutes:

68\frac{\text{kilometers}}{\text{hour}} \times \frac{1{,}000\text{ meters}}{1\text{ kilometer}} \times \frac{1\text{ hour}}{60\text{ minutes}} = \frac{68{,}000}{60}\frac{\text{meters}}{\text{minute}} \approx 1{,}133.3\text{ meters per minute}

The kilometer and hour units cancel, leaving meters per minute. The rounded result is about 1,133.3 meters per minute. This is the same rate expressed using different measurement units, not a different relationship.

When the Measurement Units Change

Changing a variable’s measurement units can change the numerical value of a slope because one unit in the new scale may represent a different amount than one unit in the old scale. The slope’s unit label must change along with its number. A sound conversion keeps the predicted response change consistent for the same real-world change in the predictor.

For example, the Celsius and Fahrenheit scales have different-sized units. A change of 1 degree Celsius corresponds to a change of 1.8 degrees Fahrenheit. If the response is measured in kilowatt-hours, a slope in kilowatt-hours per degree Celsius will have a different numerical value when expressed in kilowatt-hours per degree Fahrenheit. The predictions for a particular temperature change, however, should agree after the conversion.

Worked Example: Converting a Temperature Slope

A fictional model predicts a building’s daily electricity use $E$, in kilowatt-hours, from the outdoor temperature $C$, in degrees Celsius:

\hat{E}=84-1.6C

Find the slope units and convert the slope to kilowatt-hours per degree Fahrenheit. Check that a 5-degree Celsius increase gives the same predicted change as a 9-degree Fahrenheit increase.

The slope is $-1.6$, and its units are kilowatt-hours per degree Celsius. Its negative sign means that, according to the fitted line, predicted electricity use decreases as the outdoor temperature increases. For a 5-degree Celsius increase, the predicted change is:

(-1.6\text{ kilowatt-hours per }^\circ\text{C})(5^\circ\text{C}) =-8\text{ kilowatt-hours}

A 5-degree Celsius increase is equivalent to a 9-degree Fahrenheit increase. Therefore, the slope per degree Fahrenheit must be $-8/9$, or about $-0.8889$, kilowatt-hours per degree Fahrenheit. The unit conversion gives the same result:

-1.6\frac{\text{kilowatt-hours}}{^\circ\text{C}} \times \frac{1^\circ\text{C}}{1.8^\circ\text{F}} = -0.8889\frac{\text{kilowatt-hours}}{^\circ\text{F}} \quad\text{(rounded)}

Now use the Fahrenheit rate for a 9-degree increase:

(-0.8889\text{ kilowatt-hours per }^\circ\text{F})(9^\circ\text{F}) \approx -8.0001\text{ kilowatt-hours} \approx -8\text{ kilowatt-hours}

The tiny difference before rounding comes from using the rounded slope $-0.8889$. Both calculations give a predicted decrease of 8 kilowatt-hours for the same temperature change. In context, the Fahrenheit version says: For each 1-degree Fahrenheit increase in outdoor temperature, the line predicts daily electricity use will decrease by about 0.8889 kilowatt-hours. This describes the fitted association, not proof that temperature alone causes the change.

A Reliable Unit-Checking Routine

When you see a regression slope, use its units as a check on your interpretation. First identify which variable is the response and which is the predictor. Then put the response units over the predictor units. Finally, attach the rate to an increase of one predictor unit, and use the sign of the slope to choose “increases” or “decreases.”

Name the variables and units.
Identify what $x$ measures and its units, then identify what $y$ measures and its units.

Form response units per predictor unit.
Place the response units first and predictor units after “per.” This is the slope’s unit label.

State the predicted rate in context.
Say how much the predicted response changes for each one-unit increase in the predictor, naming both variables and units.

Check any unit conversion.
Convert the rate and its unit label together. Confirm that equivalent changes in the predictor produce the same predicted response change.

This routine builds on the slope interpretation from Tutorial 843, “Interpreting the Slope in Context.” It adds a specific unit check: the predictor’s units belong after “per,” and the response’s units describe the predicted change. If the units in your sentence do not match that structure, revisit which variable is the response and which is the predictor.

Common Mistakes and AP Exam Tips

Reversing the units. If price is the response and area is the predictor, the slope is dollars per square foot, not square feet per dollar. Put the response units first.
Leaving out “per.” Saying “the slope is 185 dollars” is incomplete when the predictor is area in square feet. State “185 dollars per square foot.”
Confusing slope with a response-to-predictor ratio. The slope is predicted response change divided by predictor change, not necessarily $y/x$. Refer to changes when explaining the rate.
Using response units alone. A slope of $-1.6$ in the temperature example is not simply $-1.6$ kilowatt-hours. It is $-1.6$ kilowatt-hours per degree Celsius.
Converting the number but not the units. A value of 68 kilometers per hour cannot be relabeled as 68 meters per minute. Convert the numerator and denominator, and show the units cancelling.
Describing every individual as if the line guarantees the change. A full-credit interpretation refers to the change in the line’s predicted response. Actual cases can lie above or below the line.

For full-credit AP communication, state the direction of change, the amount of predicted change, the one-unit increase in the predictor, and both variables in context. Include the correct response-per-predictor units. When converting units, show enough work to make clear that the converted rate describes the same change.

Key takeaway: A regression slope is a rate of change in predicted response, measured in response units per predictor unit. Keep the units attached to the number, and convert both the value and the unit label when measurement scales change.

Check Your Understanding

Use response units per predictor unit to answer each question.

A regression predicts monthly rent in dollars from apartment area in square feet. What units should the slope have?
A line predicts distance in miles from time in minutes and has slope $0.4$. State the slope units and explain what the slope means for the line’s predicted distance.
A model predicts plant height in centimeters from days since planting. The slope is $1.2$. Why would “the slope is 1.2 centimeters” be incomplete?
A slope is $30$ dollars per hour. How much does the line’s predicted response change for a 4-hour increase in the predictor?
A slope is $-2$ liters per minute. If the predictor is changed from minutes to seconds, should the slope’s numerical value stay $-2$? Explain what must happen to its units.