How to interpret the slope of a regression line

By Jude Wallis · Updated July 31, 2026

For each additional one unit of $x$, the model predicts $y$ increases (or decreases) by $b$ units, on average. Say predicted, not caused, name both variables with their units, and interpret the intercept only when $x = 0$ sits inside the data.

AP Statistics: Unit 5 (topics 5.3 Linear Regression Models, 5.5 Least-Squares Regression). Slope interpretation is Unit 5 topic 5.5 of the Fall 2026 AP Statistics course, with predicted values and extrapolation in topic 5.3; Unit 5 carries 10 to 20 percent of the multiple-choice section, and the redesigned exam no longer includes inference for regression slopes.

The sentence that earns the point

Fill in this template and you have the whole point:

For each additional [one unit of x], the model predicts [the response variable] will [increase or decrease] by [slope] [units of y], on average.

Every bracket has to carry real context. The AP course puts it this way in Unit 5: the slope of the least-squares regression line is the predicted increase or decrease in the response variable for a one-unit increase in the explanatory variable, and it should be interpreted in context.

Name the symbols before going further. The fitted line is $\hat{y} = a + bx$ , where $\hat{y}$ (say "y-hat") is the predicted value of the response variable, $x$ is the explanatory variable, $b$ is the slope, and $a$ is the y-intercept. The slope carries units of y per one unit of x, which is why the units can never be dropped.

Regression is Unit 5 of the Fall 2026 course, worth 10 to 20 percent of the multiple-choice section, and slope interpretation also turns up inside free-response questions. For how the line itself is built, see the least-squares regression line.

Why each phrase in the template is there

"For each additional one unit of x." A slope is a rate, so it needs a denominator. Naming the unit of $x$ (one more stop, one more gram, one more year) supplies it. Writing "as x goes up" without saying by how much leaves the rate undefined.

"The model predicts." Regression returns $\hat{y}$ , a predicted value, not an observed one. No individual case is required to move by exactly $b$ when $x$ rises by one. The line describes where the cloud of points is centered, so the change it reports is a change in the prediction.

"Increase or decrease by [slope]." Read the sign off the coefficient and say it in words. A slope of -5.8 is a predicted decrease of 5.8 units, not a change of "negative 5.8".

"[Units of y], on average." The number 1.5 is meaningless until it becomes 1.5 minutes. Adding "on average" makes the prediction language explicit; the AP wording anchors on "predicted", and saying both costs you nothing.

The four ways students lose this point

1. Causal language. "Adding a stop causes the trip to take 1.5 minutes longer" claims cause and effect. The AP course states plainly that a real or perceived relationship between two variables does not mean changes in one cause changes in the other, and only random assignment of treatments in an experiment supports a cause-and-effect conclusion. With observational data, write "predicts" or "is associated with" instead. The full argument is in correlation vs causation.

2. Omitting units. "The trip time increases by 1.5" leaves a grader with no idea whether that is 1.5 minutes, 1.5 hours, or 1.5 percent. Units on the response variable are part of the answer, and so is the unit attached to the one-unit increase in $x$ .

3. Omitting the prediction language. "The trip time increases by 1.5 minutes" asserts that the trip time actually does this. It is a statement about the world rather than about the model. Insert "the model predicts" or "the predicted trip time" and the same sentence becomes correct.

4. Forcing an interpretation of the intercept. Interpreting $a$ as a real quantity when $x = 0$ lies outside the data, or when $x = 0$ is impossible in context, is the fourth reliable way to lose credit. The next section handles it.

The intercept, and when to refuse to interpret it

The y-intercept $a$ is the predicted value of the response variable when the explanatory variable equals 0, stated in context. The template is: "When [x] is 0 [units], the model predicts [y] is [a] [units]."

The AP course names two situations where that sentence should not be written as though it means something. First, $x = 0$ can fall beyond the interval of x-values used to fit the line, which makes the intercept an extrapolation, and a predicted value gets less reliable the further you extrapolate. Second, the intercept can be logically impossible in context, such as a negative value for a response variable that has no negative values, like height.

So the safe procedure has two steps. Look at the range of observed x-values. If 0 sits inside that range and makes sense in context, interpret the intercept normally. If 0 sits outside the range, say so: "The intercept is 15 minutes, but the routes observed had 8 to 28 stops, so $x = 0$ is an extrapolation and the intercept has no meaningful interpretation here." That sentence earns credit; a confident interpretation of an impossible value does not.

The same reasoning governs predictions in general. Predicting inside the observed range is interpolation and is trustworthy; predicting outside it is extrapolation. See interpolation vs extrapolation.

One interpretation, graded

Take the line $\hat{y} = 15 + 1.5x$ , where $x$ is the number of stops on a bus route and $y$ is the total trip time in minutes. Here is the same interpretation at four levels of quality.

Answer	Verdict	What is wrong
"The slope is 1.5."	No credit	Restates the number. No context, no units, no direction.
"Adding a stop causes the trip to take 1.5 minutes longer."	No credit	Causal claim from observational data, and it describes the world rather than the model.
"For each additional stop, the trip time increases by 1.5."	Not full credit	Units missing, and "increases" asserts an actual change instead of a predicted one.
"For each additional stop on a route, the model predicts the total trip time increases by 1.5 minutes, on average."	Full credit	Nothing. Context, units, direction, and prediction language are all present.

The gap between the third row and the fourth is two scored elements: the words "the model predicts", and the unit minutes. That is the entire difference between a partial score and a full one, which is why the template is worth writing out until it is automatic.

Negative slopes and awkward units

A negative slope changes one word. If $b = -0.42$ and $y$ is measured in grams while $x$ is measured in days, write: "For each additional day, the model predicts the mass decreases by 0.42 grams, on average." Do not write "increases by -0.42 grams".

When the explanatory variable has awkward units, keep the one-unit phrasing and let the number be small. If $x$ is population in thousands, a one-unit increase is one thousand people, so say "for each additional thousand people". If the natural question is about a different step size, scale the slope and say you did: a slope of -5.8 pints per dollar means a predicted decrease of $5.8 \times 0.5 = 2.9$ pints for a 50-cent increase.

One more habit worth building. Before interpreting anything, confirm a line was the right model by checking that the residual plot shows no pattern. A slope interpreted off a curved relationship is a well-worded description of a bad model. See how to interpret residual plots.

What the slope does not tell you

It is not the correlation. The correlation coefficient $r$ has no units and stays between -1 and 1; the slope carries units and can be any size. They always share a sign, and they answer different questions. The side-by-side is in slope vs correlation coefficient.

It is not a measure of strength. A steep slope can come from data scattered far from the line, and a gentle slope can come from points sitting almost exactly on it. Strength is the job of $r$ and $r^2$ .

It is not evidence of causation on its own. Random assignment of treatments in an experiment is what supports a cause-and-effect conclusion, because it reduces the potential for confounding variables. A slope computed from an observational study describes an association.

It is no longer tested inferentially on the AP exam. Inference for regression slopes, meaning confidence intervals and t-tests for the true slope, was removed in the Fall 2026 redesign. It remains standard content in a college course, so you may still meet it, but the AP exam asks you to calculate and interpret the slope rather than test it. You can check any line you fit with the regression calculator.

Fit the line, then interpret the slope and the intercept

A transit agency records the number of stops on each of six weekday bus routes, $x$ , and the total trip time in minutes from the first stop to the last, $y$ : (8, 28), (12, 32), (16, 39), (20, 45), (24, 50), (28, 58). Find the least-squares regression line and interpret both coefficients in context.

Find the means. $\bar{x}$ ("x-bar") $= \frac{8+12+16+20+24+28}{6} = \frac{108}{6} = 18$ stops, and $\bar{y}$ ("y-bar") $= \frac{28+32+39+45+50+58}{6} = \frac{252}{6} = 42$ minutes.
Find each deviation from its mean. For $x$ : -10, -6, -2, 2, 6, 10. For $y$ : -14, -10, -3, 3, 8, 16.
Sum the squared x-deviations. $100 + 36 + 4 + 4 + 36 + 100 = 280$ .
Sum the products of paired deviations. $(-10)(-14) + (-6)(-10) + (-2)(-3) + (2)(3) + (6)(8) + (10)(16) = 140 + 60 + 6 + 6 + 48 + 160 = 420$ .
Find the slope. $b = \frac{420}{280} = 1.5$ minutes per stop.
Find the intercept. $a = \bar{y} - b\bar{x} = 42 - 1.5(18) = 42 - 27 = 15$ minutes.
Check the line against the mean point. At $x = 18$ , $\hat{y} = 15 + 1.5(18) = 15 + 27 = 42$ , which equals $\bar{y}$ , as every least-squares line must.
Interpret the slope. For each additional stop on a route, the model predicts the total trip time increases by 1.5 minutes, on average.
Interpret the intercept, carefully. The routes observed had 8 to 28 stops, so $x = 0$ lies outside the data and a route with no stops is not a route. Report that the intercept of 15 minutes is an extrapolation with no meaningful interpretation in this context.

The line is $\hat{y} = 15 + 1.5x$ . Slope: for each additional stop on a route, the model predicts the total trip time increases by 1.5 minutes, on average. Intercept: 15 minutes is the predicted trip time for a route with 0 stops, but 0 falls outside the observed range of 8 to 28 stops, so the intercept has no meaningful interpretation here.

Interpret a negative slope, predict, and read the residual

A farm stand fits a line to eight Saturdays of data, with $x$ the price per pint of blueberries in dollars and $y$ the number of pints sold per hour. The prices used ranged from 3.00 to 7.00 dollars and the fitted line is $\hat{y} = 42.6 - 5.8x$ . Interpret the slope, predict sales at a price of 5.50 dollars, and find the residual for a Saturday at 5.50 dollars when 12 pints per hour actually sold.

Read the slope and its sign. $b = -5.8$ pints per hour per dollar, so the predicted change is a decrease.
Interpret the slope. For each additional dollar in the price per pint, the model predicts the number of pints sold per hour decreases by 5.8, on average.
Predict at 5.50 dollars, which is interpolation since 5.50 lies between 3.00 and 7.00. $\hat{y} = 42.6 - 5.8(5.50)$ .
Do the multiplication. $5.8 \times 5.50 = 31.9$ , so $\hat{y} = 42.6 - 31.9 = 10.7$ pints per hour.
Find the residual. Residual $= y - \hat{y} = 12 - 10.7 = 1.3$ pints per hour.
Read the sign of the residual. It is positive, so the model underpredicted sales on that Saturday by 1.3 pints per hour.
Do not interpret the intercept. A price of 0 dollars is outside the observed range of 3.00 to 7.00, so the 42.6 pints per hour it predicts is an extrapolation.
Rescale the slope if a smaller step is more useful. A 50-cent increase gives a predicted decrease of $5.8 \times 0.5 = 2.9$ pints per hour.

For each additional dollar in the price per pint, the model predicts the number of pints sold per hour decreases by 5.8, on average. At 5.50 dollars, $\hat{y} = 10.7$ pints per hour, and the residual for an observed 12 pints per hour is $12 - 10.7 = 1.3$ , so the model underpredicted that Saturday.

Frequently asked questions

Do I have to write "on average" to get the point?

The AP wording anchors on "predicted", so "the model predicts the trip time increases by 1.5 minutes" is already correct. Adding "on average" makes the same idea explicit and costs nothing, so include it. What is not optional is having at least one of the two, since a sentence with neither claims an actual change rather than a predicted one.

Can I ever say the slope shows cause and effect?

Only when the data came from an experiment with random assignment of treatments, which reduces the potential for confounding variables and is what supports a cause-and-effect conclusion. For observational data, and for anything described as a survey or a set of records, use "predicts" or "is associated with".

What do I say when the intercept is negative or impossible?

Say that directly. Write that the intercept is the predicted response at $x = 0$ , that $x = 0$ is outside the interval of x-values used to fit the line or is impossible in context, and that the intercept therefore has no logical interpretation here. Naming the problem is the answer the question is looking for.

Does the AP exam still test confidence intervals and tests for the slope?

No. Inference for regression slopes was removed from the AP course in the Fall 2026 redesign. Unit 5 now asks you to calculate a predicted value, calculate the coefficients with technology, and interpret the slope, intercept, $r$ , and $r^2$ in context. Inference for slopes is still standard in college courses.