AP Statistics regression cheat sheet

By Jude Wallis · Published

Describe a scatterplot by form, direction, strength, and unusual features. The least-squares line is y^=a+bx\hat{y} = a + bx, where b=r⋅sysxb = r \cdot \frac{s_y}{s_x} and a=yˉ−bxˉa = \bar{y} - b\bar{x}. A residual is observed minus predicted, and r2r^2 is the share of variation in yy the line explains.

AP Statistics: Unit 5 (topics 5.1 Graphical Representations Between Two Quantitative Variables, 5.2 Correlation, 5.3 Linear Regression Models, 5.4 Residuals, 5.5 Least-Squares Regression). Regression is Unit 5, topics 5.1 through 5.5, of the Fall 2026 AP Statistics course. Inference for a regression slope and the old topic on analyzing departures from linearity were removed in the redesign.

Describing a scatterplot

The explanatory variable goes on the x-axis and the response variable goes on the y-axis.

FeatureWhat to stateWords to use
FormThe overall shape of the patternLinear, nonlinear (curved)
DirectionWhether yy tends to rise or fall as xx increasesPositive, negative
StrengthHow closely the points follow the patternStrong, moderate, weak
Unusual featuresPoints or groups that depart from the patternClusters, points far from the trend

Sentence pattern: "There is a [strength], [direction], [form] association between [explanatory variable] and [response variable]. As [x] increases, [y] tends to [increase or decrease]." Add any unusual features in context. "Tends to" marks the pattern as a general tendency across individuals.

Correlation

Fact about rrDetail
Range−1≤r≤1-1 \leq r \leq 1
SignPositive for a positive association, negative for a negative one; it matches the sign of the slope
StrengthCloser to −1-1 or 11 means a stronger linear association; r=0r = 0 means no linear association
UnitsNone. Changing the units of xx or yy does not change rr
What it measuresLinear association only. A strong curved relationship can have rr near 0
LinearityA value of rr near −1-1 or 11 does not by itself show that a linear model is appropriate. Check the scatterplot and the residual plot
CausationA strong correlation does not show that changes in xx cause changes in yy

Technology reports rr. See how to find r by hand for where the number comes from, and correlation vs causation.

The least-squares regression line

ItemFormula or fact
Modely^=a+bx\hat{y} = a + bx
Slopeb=r⋅sysxb = r \cdot \frac{s_y}{s_x}
Intercepta=yˉ−bxˉa = \bar{y} - b\bar{x}
CriterionThe line that makes the sum of the squared residuals as small as possible
Mean pointThe line passes through (xˉ,yˉ)(\bar{x}, \bar{y})
PredictionSubstitute a value of xx to get y^\hat{y}
ExtrapolationPredicting for an xx outside the range of the data; the prediction becomes less reliable the further outside it is

y^\hat{y} ("y-hat") is the predicted value of the response, aa is the y-intercept, bb is the slope, xˉ\bar{x} and yˉ\bar{y} are the sample means, and sxs_x and sys_y are the sample standard deviations. In practice aa and bb come from technology, such as the regression calculator. See least-squares regression line.

Interpretation sentences

QuantitySentence pattern
Slope bb"For each additional 1 [unit of x], the predicted [y] [increases or decreases] by [b] [units of y]."
Intercept aa"When [x] is 0 [units], the predicted [y] is [a] [units of y]."
Correlation rr"There is a [strength], [direction] linear association between [x] and [y]."
Coefficient of determination r2r^2"About [r2r^2 as a percent] of the variation in [y] is explained by the linear relationship with [x]."
Residual"The actual [y] was [amount] [units] [above or below] the value predicted by the line."
Prediction"The model predicts a [y] of [value with units] when [x] is [value]."

Three cautions:

  • Write "predicted" in the slope and intercept sentences. The line gives a predicted value, and individual observations vary around it.
  • The intercept may have no reasonable meaning when x=0x = 0 is far outside the data or when the value is impossible in context.
  • r2r^2 is the square of rr. To recover rr from r2r^2, take the square root and attach the sign of the slope.

See how to interpret the slope of a regression line and r vs r-squared.

Residuals and residual plots

residual=y−y^=observed−predicted\text{residual} = y - \hat{y} = \text{observed} - \text{predicted}
ResidualPosition of the pointThe line
PositiveAbove the lineUnderpredicted the response
NegativeBelow the lineOverpredicted the response
ZeroOn the linePredicted the response exactly

A residual plot graphs the residuals against the explanatory variable or against the predicted values.

Residual plot showsConclusion
Points scattered randomly above and below 0, with no patternA linear model is appropriate
A curved patternA linear model is not appropriate

A residual has the units of the response variable. See how to interpret residual plots.

What the Fall 2026 framework includes

Unit 5 has five topics: scatterplots (5.1), correlation (5.2), linear regression models and prediction (5.3), residuals (5.4), and least-squares regression (5.5). Everything in the sections above belongs to those topics.

The College Board's list of topics removed for the Fall 2026 course affects regression in two ways:

  • Inference for the slope of a regression line (the old Unit 9) was removed. The course has no confidence interval and no significance test for a slope.
  • The old topic on analyzing departures from linearity was removed, so the course has no topic on transforming data to achieve linearity. Curvature in a residual plot is still used as evidence that a linear model is not appropriate.

See what was removed from AP Statistics and Unit 5.

Build the line from summary statistics and find a residual

For 30 students, hours studied (xx) has mean 5 and standard deviation 2, exam score (yy) has mean 70 and standard deviation 8, and the correlation is r=0.75r = 0.75. The scatterplot is linear. Find the least-squares line, predict the score for a student who studied 6 hours, find the residual for a student who studied 6 hours and scored 78, and interpret r2r^2.

  1. Slope: b=r⋅sysx=0.75⋅82=3b = r \cdot \frac{s_y}{s_x} = 0.75 \cdot \frac{8}{2} = 3.

  2. Intercept: a=yˉ−bxˉ=70−3(5)=55a = \bar{y} - b\bar{x} = 70 - 3(5) = 55.

  3. Line: y^=55+3x\hat{y} = 55 + 3x.

  4. Prediction at x=6x = 6: y^=55+3(6)=73\hat{y} = 55 + 3(6) = 73 points.

  5. Residual: y−y^=78−73=5y - \hat{y} = 78 - 73 = 5 points. The residual is positive, so the line underpredicted this student's score.

  6. Coefficient of determination: r2=0.752=0.5625r^2 = 0.75^2 = 0.5625.

The line is y^=55+3x\hat{y} = 55 + 3x. For each additional hour studied, the predicted exam score increases by 3 points. The predicted score at 6 hours is 73, the residual for the student who scored 78 is 5 points, and about 56% of the variation in exam score is explained by the linear relationship with hours studied.

Interpret a fitted line and spot extrapolation

A least-squares line for predicting the price of a used car of one model, in thousands of dollars, from its age in years is y^=24.0−1.8x\hat{y} = 24.0 - 1.8x. It was fit to cars from 1 to 10 years old, and r2=0.81r^2 = 0.81. Interpret the slope, find rr, find the residual for a 5-year-old car that sold for 13.6 thousand dollars, and comment on a prediction for a 15-year-old car.

  1. Slope: for each additional year of age, the predicted price decreases by 1.8 thousand dollars.

  2. Correlation: 0.81=0.9\sqrt{0.81} = 0.9, and the slope is negative, so r=−0.9r = -0.9.

  3. Prediction at x=5x = 5: y^=24.0−1.8(5)=15.0\hat{y} = 24.0 - 1.8(5) = 15.0 thousand dollars.

  4. Residual: 13.6−15.0=−1.413.6 - 15.0 = -1.4 thousand dollars. The residual is negative, so the line overpredicted this car's price.

  5. At x=15x = 15: y^=24.0−1.8(15)=−3.0\hat{y} = 24.0 - 1.8(15) = -3.0 thousand dollars, a negative price. Age 15 is outside the 1 to 10 years used to fit the line, so this is extrapolation.

Slope: the predicted price drops by 1.8 thousand dollars per year of age. r=−0.9r = -0.9. The 5-year-old car's residual is −1.4-1.4 thousand dollars. The prediction for a 15-year-old car is an extrapolation and gives an impossible negative price, so it should not be used.

Frequently asked questions

What is the difference between r and r-squared?

rr gives the direction and strength of a linear association and lies between −1-1 and 11. r2r^2 is the proportion of the variation in the response that the linear relationship with the explanatory variable explains, and it lies between 0 and 1.

Does a high correlation mean a line is the right model?

No. Data that follow a curve can still have rr close to 1. The form is judged from the scatterplot and the residual plot. A curved pattern in the residual plot shows that a linear model is not appropriate.

Is inference for a regression slope part of the course?

No. The College Board lists inference for slopes among the topics removed for the Fall 2026 course. Unit 5 describes and interprets the relationship without a confidence interval or a significance test for the slope.