6.4 Two-Variable Data: Scatterplots & Line of Best Fit

Q: A marine biologist models the relationship between water depth d (in meters) and water temperature T (in degrees Celsius) in a lake using the linear regression equation T = -0.65d + 24.8, valid for 0 <= d <= 30. What is the best interpretation of the number -0.65 in this context?

For each 1-meter increase in depth, the predicted water temperature decreases by 0.65°C.. In the linear regression model T = -0.65d + 24.8, the slope is -0.65. The slope represents the estimated or predicted change in the dependent variable (temperature T in °C) for each 1-unit increase in the independent variable (depth d in meters). Therefore, for every 1-meter increase in depth, the predicted water temperature decreases by 0.65°C.

Q: A regression line ŷ = 2.4x + 15 is used to estimate the height ŷ (in inches) of sapling trees based on their age x (in months). A specific sapling tree is 10 months old and has an actual measured height of 43 inches. What is the residual for this tree, and what does it indicate about the regression model's prediction?

+4 inches; the model underestimated the actual tree height.. To compute the residual, first calculate the model's predicted height at x = 10: ŷ = 2.4(10) + 15 = 39 inches. Then evaluate Residual = y_actual - ŷ_predicted = 43 - 39 = +4 inches. Because the residual is positive, the actual measurement was greater than the modeled prediction, meaning the model underestimated the actual height.

Q: A scatterplot displays the relationship between the radius r of a circular solar panel (in meters) and the total electrical power output P (in watts). The data points closely follow the curve P = 450 r². A researcher fits a simple linear line of best fit P_hat = mr + b to the data for r between 1 and 8 meters. Which statement accurately describes the limitations of this linear model?

The linear model will systematically underestimate power output at extreme low and high values of r and overestimate power in the middle range due to the quadratic curvature of the true relationship.. When a straight linear regression line is fitted across a concave-upward quadratic dataset (P = 450 r²), the linear fit acts like a secant line: it sits above the true curve in the middle range (overestimating power) and drops below the true accelerating curve at the lower and upper extremities (underestimating power). This systematic bias illustrates why linear models are inadequate for nonlinear relationships.

Key Takeaways

Scatterplots display bivariate data pairs $(x, y)$ to uncover directional associations (positive, negative, or none) and morphological forms (linear vs. nonlinear).
In a linear regression model $\hat{y} = mx + b$, the slope $m$ represents the *predicted change* in the response variable $y$ for every 1-unit increase in the explanatory variable $x$.
The $y$-intercept $b$ represents the predicted value of $y$ when $x = 0$; always assess whether $x = 0$ is physically meaningful or represents an unobservable extrapolation.
Residuals quantify prediction error: $\text{Residual} = y_{\text{actual}} - \hat{y}_{\text{predicted}}$. Points located above the regression line have positive residuals (underestimated), while points below have negative residuals (overestimated).
Correlation does not imply causation: a strong linear association ($r \approx \pm 1$) confirms mathematical co-variation, but observational studies cannot rule out confounding variables without controlled experimentation.

Last updated: August 2026

6.4 Two-Variable Data: Scatterplots & Line of Best Fit

Two-variable (bivariate) data analysis investigates the relationship between an independent (explanatory) variable $x$ and a dependent (response) variable $y$. On the Digital SAT, questions focus on extracting quantitative models from scatterplots, interpreting regression parameters in scientific/business contexts, calculating residuals, and avoiding false causal claims.

1. Scatterplot Association Patterns

+-----------------------------------------------------------------------------+
|                        SCATTERPLOT PATTERN TYPOLOGY                         |
|                                                                             |
|   [ POSITIVE LINEAR ]       [ NEGATIVE LINEAR ]       [ NO ASSOCIATION ]    |
|   y ^        *              y ^  *                    y ^    *   *   *      |
|     |      * *                |    * *                  |  *   *   *   *    |
|     |    * *                  |      * *                |    *   *   *      |
|     |  * *                    |        * *              |  *   *   *   *    |
|     +-------------> x         +-------------> x         +-------------> x   |
|     Slope m > 0               Slope m < 0               Slope m ≈ 0         |
|     (Both increase)           (y decreases as x ↑)      (No trend)          |
|                                                                             |
|   [ QUADRATIC / NONLINEAR ]   [ EXPONENTIAL GROWTH ]                        |
|   y ^       * *             y ^            *                                |
|     |     *     *             |          *                                  |
|     |   *         *           |       *                                     |
|     | *             *         | *  *                                        |
|     +-------------> x         +-------------> x                             |
|     Curved (Parabolic)        Accelerating Growth Rate                      |
+-----------------------------------------------------------------------------+

2. Line of Best Fit (Linear Regression Equation)

A line of best fit (trendline) models the overall trajectory of bivariate data. Statistically, it is generated via the least-squares criterion, which minimizes the sum of squared vertical distances between data points and the line.

$\hat{y} = mx + b$

where:

$\hat{y}$ ("$y$-hat") represents the predicted or estimated value of the response variable.
$m$ is the slope (rate of change).
$b$ is the $y$-intercept (initial baseline value).

3. Interpreting Slope and Intercept in Context (The SAT Template)

The College Board tests precise linguistic interpretation of slope and intercept.

+-----------------------------------------------------------------------------+
|                     REGRESSION PARAMETER INTERPRETATION                     |
|                                                                             |
|   [ SLOPE (m) TEMPLATE ]:                                                   |
|   "For each increase of 1 [unit of x], the model PREDICTS / ESTIMATES       |
|    that [y-variable] will increase (if m > 0) or decrease (if m < 0)        |
|    by |m| [units of y]."                                                    |
|                                                                             |
|   [ y-INTERCEPT (b) TEMPLATE ]:                                             |
|   "The PREDICTED value of [y-variable] when [x-variable] is equal to 0."    |
|                                                                             |
|   MANDATORY WORDS: "Predicted", "Estimated", "On average", "Model"          |
|   FORBIDDEN WORDS: "Always", "Exactly", "Causes", "Guarantees"             |
+-----------------------------------------------------------------------------+

Worked Example 1: Fuel Efficiency Model

Problem: An automotive engineer models highway fuel consumption for an SUV using the equation $\hat{y} = -0.04x + 38$, where $\hat{y}$ is the estimated fuel efficiency in miles per gallon (mpg) and $x$ is the vehicle speed in miles per hour (mph), valid for speeds between $45\text{ mph}$ and $85\text{ mph}$. What is the best interpretation of $-0.04$ in this context?

Step-by-Step Analysis:

Identify the parameter: The slope is $m = -0.04$.
Identify units: $x$ is in mph, $y$ is in mpg.
Apply the slope template: For each $1\text{ mph}$ increase in speed, the model predicts that fuel efficiency decreases by $0.04\text{ mpg}$.

4. Residuals: Definition, Formula & Geometry

A residual measures the vertical error between an actual observed data point and the model's predicted value.

$\text{Residual} = y_{\text{actual}} - \hat{y}_{\text{predicted}}$

+-----------------------------------------------------------------------------+
|                        RESIDUAL GEOMETRY VISUALIZATION                      |
|                                                                             |
|   y ^                                                                       |
|     |             * (Actual Point: y_actual)                                |
|     |             |                                                         |
|     |             |  <-- Positive Residual (+): Actual > Predicted          |
|     |             |      (Model UNDERESTIMATED the real value)              |
|     |          ---•--------- (Line of Best Fit: ŷ_predicted)                |
|     |             |                                                         |
|     |             |  <-- Negative Residual (-): Actual < Predicted          |
|     |             * (Actual Point: y_actual)                                |
|     |                 (Model OVERESTIMATED the real value)                  |
|     +----------------------------------------> x                            |
+-----------------------------------------------------------------------------+

Residual Sign	Point Position	Actual vs. Predicted	Model Interpretation
Positive Residual ($> 0$)	Point is above the line	$y_{\text{actual}} > \hat{y}_{\text{predicted}}$	Model underestimated the real outcome
Negative Residual ($< 0$)	Point is below the line	$y_{\text{actual}} < \hat{y}_{\text{predicted}}$	Model overestimated the real outcome
Zero Residual ($= 0$)	Point is on the line	$y_{\text{actual}} = \hat{y}_{\text{predicted}}$	Model perfectly predicted the outcome

Worked Example 2: Calculating a Residual

Problem: A researcher models sapling tree height $\hat{y}$ (in inches) based on age $x$ (in months) using $\hat{y} = 2.4x + 15$. A specific sapling is 10 months old and has an actual measured height of $43\text{ inches}$. What is the residual for this sapling?

Step-by-Step Solution:

Find the predicted value $\hat{y}$ at $x = 10$: $\hat{y} = 2.4(10) + 15 = 24 + 15 = 39\text{ inches}$
Identify the actual observed value $y$: $y_{\text{actual}} = 43\text{ inches}$
Apply the residual formula: $\text{Residual} = y_{\text{actual}} - \hat{y}_{\text{predicted}} = 43 - 39 = +4\text{ inches}$
Interpretation: The actual sapling was $4\text{ inches}$ taller than predicted; the model underestimated its height.

5. Nonlinear Scatterplot Trajectories

When scatterplot points exhibit clear curvature, fitting a linear regression line introduces systematic bias.

Quadratic Models ($\hat{y} = ax^2 + bx + c$): Appropriate when data exhibits an inflection point (projectile heights, revenue optimization).
Exponential Models ($\hat{y} = a \cdot b^x$): Appropriate when values grow or decay by a constant percentage rate per unit increase in $x$ (bacterial growth, radioactive decay, compound interest).

6. Correlation vs. Causation & Confounding Variables

+-----------------------------------------------------------------------------+
|                     THE CORRELATION ≠ CAUSATION PRINCIPLE                   |
|                                                                             |
|   [ OBSERVED CORRELATION ]                 [ THIRD CONFOUNDING VARIABLE ]   |
|   Ice Cream Sales (x)                      High Summer Temperatures (Z)     |
|          |                                            /           \         |
|          v                                           v             v        |
|   Drowning Incidents (y)                   Ice Cream (x)     Drownings (y)  |
|                                                                             |
|   Ice cream sales do NOT cause drownings. Both are driven by hot weather!   |
+-----------------------------------------------------------------------------+

Correlation: Measures the statistical strength and direction of linear association between two variables ($-1 \le r \le 1$).
Causation: Proves that changes in variable $x$ directly produce changes in variable $y$.
The Golden SAT Rule: An observational scatterplot can never establish causation on its own, regardless of how strong the correlation is ($r = 0.99$). Only a randomized controlled experiment can establish causality.

7. Desmos Playbook for Scatterplots, Regressions & Residuals

Desmos offers a powerful regression suite that fits linear, polynomial, and exponential curves to coordinate data in seconds.

+-----------------------------------------------------------------------------+
|                        DESMOS REGRESSION & RESIDUAL PLAYBOOK                |
|                                                                             |
|   1. Create Coordinate Table:                                               |
|      Click [ + ] -> Select Table -> Enter x1 and y1 coordinates             |
|                                                                             |
|   2. Execute Regression Models on Next Line:                                |
|      Linear Model:         y1 ~ m x1 + b                                    |
|      Quadratic Model:      y1 ~ a x1^2 + b x1 + c                           |
|      Exponential Model:    y1 ~ a b^(x1)                                    |
|                                                                             |
|   3. Inspect Regression Outputs:                                            |
|      - Parameters: Exact values for m, b, a, c                              |
|      - Statistics: r (correlation coefficient), r^2 (variance explained)    |
|      - Residuals Plot: Click 'Plot' next to e1 to graph residuals     |
+-----------------------------------------------------------------------------+

[!TIP] Rapid Regression Verification in Desmos: When an SAT problem provides three or four $(x, y)$ data points and asks which equation best fits the data, enter the points into a Desmos table and type y1 ~ m x1 + b. Desmos outputs the exact slope and $y$-intercept instantly, saving 2+ minutes of manual slope and point-slope arithmetic.

Test Your Knowledge

A marine biologist models the relationship between water depth d (in meters) and water temperature T (in degrees Celsius) in a lake using the linear regression equation T = -0.65d + 24.8, valid for 0 <= d <= 30. What is the best interpretation of the number -0.65 in this context?

The predicted water temperature at the surface of the lake is -0.65°C.

The maximum depth of the lake is estimated to decrease by 0.65 meters per day.

For each 1-meter increase in depth, the predicted water temperature decreases by 0.65°C.

The water temperature decreases by 24.8°C for every 0.65 meters of depth.

Test Your Knowledge

A regression line ŷ = 2.4x + 15 is used to estimate the height ŷ (in inches) of sapling trees based on their age x (in months). A specific sapling tree is 10 months old and has an actual measured height of 43 inches. What is the residual for this tree, and what does it indicate about the regression model's prediction?

+4 inches; the model underestimated the actual tree height.

-4 inches; the model overestimated the actual tree height.

+4 inches; the model overestimated the actual tree height.

+39 inches; the model perfectly predicted the actual tree height.

Test Your Knowledge

A scatterplot displays the relationship between the radius r of a circular solar panel (in meters) and the total electrical power output P (in watts). The data points closely follow the curve P = 450 r². A researcher fits a simple linear line of best fit P_hat = mr + b to the data for r between 1 and 8 meters. Which statement accurately describes the limitations of this linear model?

The linear model will have a negative slope because power output increases with radius.

The linear model is perfectly accurate because all data points lie on the curve.

The linear model can establish that increasing radius causes power increases in all solar panel designs.

The linear model will systematically underestimate power output at extreme low and high values of r and overestimate power in the middle range due to the quadratic curvature of the true relationship.

Up Next

6.5 Probability & Two-Way Frequency Tables

Continue learning

SAT Math

SAT Math

6.4 Two-Variable Data: Scatterplots & Line of Best Fit

Key Takeaways

6.4 Two-Variable Data: Scatterplots & Line of Best Fit

1. Scatterplot Association Patterns

2. Line of Best Fit (Linear Regression Equation)

3. Interpreting Slope and Intercept in Context (The SAT Template)

Worked Example 1: Fuel Efficiency Model

4. Residuals: Definition, Formula & Geometry

Worked Example 2: Calculating a Residual

5. Nonlinear Scatterplot Trajectories

6. Correlation vs. Causation & Confounding Variables

7. Desmos Playbook for Scatterplots, Regressions & Residuals

SAT Math

1Chapter 1: Introduction & Digital SAT Math Mechanics

2Chapter 2: Algebra — Linear Equations, Functions & Inequalities

3Chapter 3: Algebra — Systems of Linear Equations & Applications

4Chapter 4: Advanced Math — Quadratics & Nonlinear Equations

5Chapter 5: Advanced Math — Exponentials, Polynomials & Functions

6Chapter 6: Problem-Solving and Data Analysis

7Chapter 7: Geometry and Trigonometry

SAT Math

6.4 Two-Variable Data: Scatterplots & Line of Best Fit

Key Takeaways

6.4 Two-Variable Data: Scatterplots & Line of Best Fit

1. Scatterplot Association Patterns

2. Line of Best Fit (Linear Regression Equation)

3. Interpreting Slope and Intercept in Context (The SAT Template)

Worked Example 1: Fuel Efficiency Model

4. Residuals: Definition, Formula & Geometry

Worked Example 2: Calculating a Residual

5. Nonlinear Scatterplot Trajectories

6. Correlation vs. Causation & Confounding Variables

7. Desmos Playbook for Scatterplots, Regressions & Residuals