01 / 24

Interpreting Scatterplots
and Choosing a Model

Learn to read scatterplots, identify patterns in data, and select the best regression model: linear, quadratic, or exponential.

College Algebra

What Is a Scatterplot?

A scatterplot is a graph of ordered pairs \((x, y)\) plotted on a coordinate plane. Each dot represents one data observation. Scatterplots reveal patterns, trends, and outliers that help us decide which mathematical model fits best.

Independent Variable (x) Response (y) outlier

Key Vocabulary

Correlation
The strength and direction of a linear relationship between two variables.
Regression
A method for fitting a curve (line, parabola, etc.) to data to make predictions.
Residual
The difference between the observed \(y\) and the predicted \(\hat{y}\): \(\text{residual} = y - \hat{y}\).

Why Does Choosing the Right Model Matter?

Forcing a straight line through curved data produces systematic errors. The residuals will show a clear pattern instead of random scatter: predictions far from the data at the extremes and too close in the middle.

bad fit good fit

Three Core Regression Models

Linear
\( y = ax + b \)
Constant rate of change. Data rises or falls in a straight line.
Quadratic
\( y = ax^2 + bx + c \)
Data curves with a single peak or valley (vertex).
Exponential
\( y = ab^x \)
Data grows or decays by a constant factor, not a constant amount.

Recognizing a Linear Pattern

x y \(y = ax + b\)

Recognizing a Quadratic Pattern

x y \(y = ax^2 + bx + c\)

Recognizing an Exponential Pattern

x y \(y = ab^x\)

Differences and Ratios: The Decision Rules

Test If Roughly Constant... Choose
First differences \(\Delta y\) Constant Linear
Second differences \(\Delta^2 y\) Constant Quadratic
Consecutive ratios \(\frac{y_{n+1}}{y_n}\) Constant Exponential

The Correlation Coefficient \(r\)

For a linear model, the value \(r\) (Pearson's correlation) measures how tightly data clusters around the line. Its square, \(r^2\), tells us the proportion of variance explained by the model.

\(r\) near \(+1\)
Strong positive linear relationship: as \(x\) increases, \(y\) increases.
\(r\) near \(-1\)
Strong negative linear relationship: as \(x\) increases, \(y\) decreases.
\(r\) near \(0\)
No linear association. But data may still have a nonlinear pattern!

Reading Residual Plots

Good: Random Scatter Bad: U-Shape Pattern x x

Linear vs Exponential: The Critical Difference

Feature Linear \(y = ax + b\) Exponential \(y = ab^x\)
Growth type Constant amount Constant factor
Rate of change Slope \(a\) is fixed Rate increases as \(y\) increases
Test First differences constant Consecutive ratios constant
Long-term Grows without bound (slowly) Grows without bound (rapidly)
Graph shape Straight line J-curve or decay curve

How to Choose the Right Model

1
Plot the data. Create a scatterplot and look at the overall shape: straight, curved up/down, or J-shaped.
2
Compute differences and ratios. For equally spaced \(x\), check first differences, second differences, and consecutive ratios.
3
Fit the candidate model. Use regression (calculator or software) to obtain the equation and \(r^2\).
4
Examine residuals. Plot residuals vs. \(x\). Random scatter confirms a good fit; a pattern signals the wrong model.
5
Compare \(r^2\) values. Among models with random residuals, pick the one with the highest \(r^2\).

Example 1: Linear Data

A student measures the distance (km) a car travels over time (hours). The data: \((1, 60), (2, 122), (3, 181), (4, 240), (5, 301)\).

Hours Distance (km) 1 2 3 4 5 60 150 240 330

Example 2: Quadratic Data

A ball is launched upward. Height (m) is recorded at times (s): \((0, 2), (1, 27), (2, 42), (3, 47), (4, 42), (5, 27), (6, 2)\).

Time (s) Height (m) 0 1 2 3 4 5 6

Example 3: Exponential Data

A bacteria colony doubles every hour. Population counts: \((0, 50), (1, 100), (2, 200), (3, 400), (4, 800), (5, 1600)\).

Hours Population 0 1 2 3 4 5 0 800 1600

Common Mistakes

✕ Always Using Linear
Students default to a straight line without checking if the data curves. Always plot first and inspect the shape.
✕ Ignoring Residuals
A high \(r^2\) does not guarantee a good fit. A patterned residual plot reveals the model is wrong.
✕ Confusing \(r\) With \(r^2\)
\(r = 0.9\) means \(r^2 = 0.81\), so only 81% of variance is explained: not 90%. Always square to interpret proportion.

Model Comparison Summary

Property Linear Quadratic Exponential
Equation form \(y = ax + b\) \(y = ax^2 + bx + c\) \(y = ab^x\)
Shape Straight line Parabola (U or ∩) J-curve or decay
Key test \(\Delta y\) constant \(\Delta^2 y\) constant Ratios constant
Direction changes 0 1 (vertex) 0
Real-world example Constant speed travel Projectile motion Bacterial growth

Putting It All Together

Plot the Scatterplot What shape do you see? Straight Line
Try \(y = ax + b\)
U or ∩ Shape
Try \(y = ax^2 + bx + c\)
J-Curve / Decay
Try \(y = ab^x\)
Check residuals and compare r²

Question 1 of 5

A dataset has equally spaced \(x\)-values. The first differences of \(y\) are: 12, 12, 11, 13, 12. Which model is most appropriate?

Question 2 of 5

After fitting a linear regression, the residual plot shows a clear U-shape. What should you conclude?

Question 3 of 5

A population triples every year: 10, 30, 90, 270, 810. What is the consecutive ratio, and which model fits?

Question 4 of 5

A scatterplot shows data that rises, reaches a maximum, then falls. Which regression model should you try first?

Question 5 of 5

Two models both produce residual plots with random scatter. Model A has \(r^2 = 0.87\) and Model B has \(r^2 = 0.95\). Which should you choose?