4.8Fitting Exponential Models to Data
Use linearization and regression to find the best-fit exponential or logarithmic model for real data. Apply the two-point method, interpret R², and choose between model types based on data patterns.
Data scientists, epidemiologists, economists, and engineers all fit exponential models to real-world data. Understanding linearization and R² is the foundation of statistical modeling — skills that appear in AP Statistics, college courses, and every data-driven career.
Essential Question
When real-world data doesn't follow a perfect formula, how do we find the exponential or logarithmic model that best fits the data — and how do we know how good that fit is?
Lesson Overview
Real data is messy. Rather than finding an exact formula, we find the best-fit model — the curve that minimizes the total error between predicted and actual values. For exponential data (y = a·bˣ), the key technique is linearization: take ln(y) to transform the curved data into a straight line, then apply linear regression to that transformed data. The slope of the line gives ln(b) and the y-intercept gives ln(a). When only two data points are available, the two-point method gives an exact model through those points. The coefficient of determination R² measures how well the model fits: R² = 1 is a perfect fit; R² close to 0 means the model explains little of the variation.
Linearization Steps (Exponential)
- Compute Y = ln(y) for each data point
- Perform linear regression on (x, Y) pairs
- Get slope m = ln(b) and intercept c = ln(a)
- Recover: b = eᵐ and a = eᶜ
- Write model: y = a · bˣ
Linearization Steps (Logarithmic)
- Compute X = ln(x) for each data point
- Perform linear regression on (X, y) pairs
- Get slope b and intercept a directly
- Write model: y = a + b·ln(x)
- Check: y vs ln(x) should look linear
Worked Examples
Use the two-point method to find an exponential model y = a·bˣ passing through (1, 6) and (4, 48).
Write two equations: 6 = a·b¹ and 48 = a·b⁴
Divide: 48/6 = b⁴/b¹ → 8 = b³
Solve for b: b = 8^(1/3) = 2
Solve for a: 6 = a·2¹ → a = 3
Model: y = 3·2ˣ
Verify: y(1) = 3·2 = 6 ✓ y(4) = 3·16 = 48 ✓
The table shows bacteria counts (thousands) at hours x = 0, 1, 2, 3: y = 2.1, 3.2, 4.8, 7.3. Use linearization to find the best-fit exponential model.
Compute Y = ln(y): ln(2.1)≈0.742, ln(3.2)≈1.163, ln(4.8)≈1.569, ln(7.3)≈1.988
Linear regression on (x, Y): slope m ≈ (1.988−0.742)/(3−0) ≈ 0.415
Y-intercept c ≈ 0.742 − 0.415·0 = 0.742 (using point (0, 0.742))
Recover: b = e^(0.415) ≈ 1.515 and a = e^(0.742) ≈ 2.100
Model: y ≈ 2.10 · (1.515)ˣ
Check: y(2) = 2.10·(1.515)² ≈ 2.10·2.295 ≈ 4.82 ≈ 4.8 ✓
A dataset gives (x, y): (1, 5.2), (2, 6.8), (4, 8.9), (8, 11.4), (16, 14.1). Determine whether an exponential or logarithmic model is more appropriate.
Test exponential: compute Y = ln(y): 1.649, 1.917, 2.186, 2.434, 2.646
Plot Y vs x: the values 1.649, 1.917, 2.186, 2.434, 2.646 at x = 1,2,4,8,16 — not evenly spaced, curve bends → not linear → poor exponential fit
Test logarithmic: compute X = ln(x): 0, 0.693, 1.386, 2.079, 2.773
Plot y vs X: 5.2, 6.8, 8.9, 11.4, 14.1 at X = 0, 0.693, 1.386, 2.079, 2.773 — roughly linear!
Linear regression on (X, y): slope b ≈ (14.1−5.2)/(2.773−0) ≈ 3.21, intercept a ≈ 5.2
Logarithmic model: y ≈ 5.2 + 3.21·ln(x)
An exponential regression gives y = 4.2·(0.87)ˣ with R² = 0.982. Interpret the model and R² value.
Base b = 0.87 < 1, so this is exponential decay.
The quantity decreases by about 13% per unit increase in x (since 1 − 0.87 = 0.13).
Initial value (x = 0): y = 4.2·(0.87)⁰ = 4.2
R² = 0.982 means 98.2% of the variation in y is explained by the model.
This is an excellent fit — the exponential model is highly appropriate for this data.
Use the two-point method to find a logarithmic model y = a + b·ln(x) passing through (1, 3) and (e², 9).
Write two equations: 3 = a + b·ln(1) = a + b·0 = a → a = 3
Second equation: 9 = 3 + b·ln(e²) = 3 + b·2 → 6 = 2b → b = 3
Model: y = 3 + 3·ln(x)
Verify: y(1) = 3 + 3·0 = 3 ✓ y(e²) = 3 + 3·2 = 9 ✓
Scatter plot with fitted curve y = 3.0·(1.50)ˣ — data points cluster tightly around the curve
Guided Practice
Use the two-point method to find y = a·bˣ through (0, 5) and (3, 40).
Hint: With x = 0: y = a·b⁰ = a, so a = 5 directly. Then use (3, 40) to find b.
Data: (0, 1.0), (1, 2.7), (2, 7.4), (3, 20.1). Compute Y = ln(y) for each point and identify the slope and y-intercept of the linearized data.
Hint: ln(1.0)=0, ln(2.7)≈0.993, ln(7.4)≈2.001, ln(20.1)≈3.000. The Y values are nearly 0, 1, 2, 3 — what does that tell you about the slope?
An exponential model y = 12·(0.75)ˣ has R² = 0.94. (a) Is this growth or decay? (b) What is the percent decrease per unit? (c) How good is the fit?
Hint: b = 0.75 < 1 means decay. Percent decrease = (1 − b)·100%. R² = 0.94 falls in the 'good' range.
Find a logarithmic model y = a + b·ln(x) through (e, 7) and (e³, 13).
Hint: ln(e) = 1 and ln(e³) = 3. Write two equations and solve the system for a and b.
A dataset has these (x, ln y) values: (1, 1.2), (2, 1.9), (3, 2.6), (4, 3.3). Find the exponential model y = a·bˣ.
Hint: The linearized data is already given. Find slope m and intercept c of the line through (x, ln y). Then b = eᵐ and a = eᶜ.
Key Vocabulary
Regression
A statistical method for finding the curve of best fit through a set of data points by minimizing the sum of squared errors between predicted and actual values.
Exponential Regression
Finding the best-fit model of the form y = a·bˣ for a dataset. Performed by linearizing with Y = ln(y) and applying linear regression to (x, Y).
Logarithmic Regression
Finding the best-fit model of the form y = a + b·ln(x). Performed by substituting X = ln(x) and applying linear regression to (X, y).
Linearization
Transforming a nonlinear relationship into a linear one by applying a function (such as ln) to one or both variables. Makes regression easier to perform and interpret.
Coefficient of Determination (R²)
A value between 0 and 1 measuring how well the regression model explains the variation in the data. R² = 1 is a perfect fit; R² = 0 means no explanatory power.
Two-Point Method
Using exactly two data points to determine the constants a and b in y = a·bˣ or y = a + b·ln(x). Gives an exact model through those two points.
Residual
The difference between an observed y-value and the value predicted by the model: residual = y_observed − y_predicted. Small residuals indicate a good fit.
Extrapolation
Using a model to predict values outside the range of the original data. Exponential models can give unreliable extrapolations — always note the domain of validity.
Check Your Understanding
Interactive Practice — 5 Questions
An exponential model y = a·bˣ passes through (0, 7) and (2, 63). What is b?
After linearizing exponential data, the slope of the line through (x, ln y) is 0.5 and the y-intercept is 1.2. What is b?
An exponential regression gives R² = 0.73. How would you describe the fit?
Which transformation linearizes the model y = a + b·ln(x)?
A model y = 5·(1.3)ˣ is used to predict y at x = 20. The original data only went up to x = 8. This prediction is an example of:
Independent Practice
Independent Practice
Find y = a·bˣ through (2, 12) and (5, 96) using the two-point method.
Data: (0, 2.0), (1, 5.4), (2, 14.8), (3, 40.2). Compute Y = ln(y), find the linearized slope and intercept, and write the exponential model.
An exponential model y = 8·(0.6)ˣ fits data with R² = 0.91. (a) Is this growth or decay? (b) What percent change per unit? (c) Interpret R².
Find y = a + b·ln(x) through (1, 4) and (e⁴, 16).
Data: (1, 3.1), (2, 4.4), (4, 5.7), (8, 7.0), (16, 8.3). Test both exponential and logarithmic linearization. Which model fits better? Explain.
ChallengeCommon Mistakes
Recovering a and b from linearization by using a = m (slope) and b = c (intercept) directly.
The slope m = ln(b) and intercept c = ln(a). You must exponentiate: b = eᵐ and a = eᶜ. Never skip this step.
Applying the two-point method and forgetting to verify the model with both original points.
Always substitute both original data points back into your model to confirm it passes through them exactly.
Concluding that R² = 0.85 means the model is "85% accurate" or "correct 85% of the time."
R² = 0.85 means the model explains 85% of the variation in y. It says nothing about prediction accuracy for individual points.
Using an exponential model for data that grows rapidly at first and then levels off (like a learning curve).
Rapid-then-slow growth with no upper bound suggests a logarithmic model. Rapid-then-slow growth with an upper bound suggests a logistic model.
Math Tips
Quick check for exponential data: compute ratios y₂/y₁, y₃/y₂, y₄/y₃. If the ratios are roughly constant, the data is exponential.
Quick check for logarithmic data: compute differences y₂−y₁, y₃−y₂ for equally spaced ln(x) values. If differences are roughly constant, the data is logarithmic.
When x = 0 is in the dataset, the two-point method is easiest: a = y(0) directly, then solve for b using the second point.
Extrapolation with exponential models is risky. Exponential growth is unbounded — real-world constraints (resources, space, competition) eventually slow or stop growth.
R² closer to 1 is better, but context matters. In biology R² = 0.85 may be excellent; in physics R² = 0.99 may be expected. Always compare R² values across candidate models.