Eighth Grade Mathematics: Making Predictions: Interpolation vs. Extrapolation
The ultimate goal of constructing a linear model from bivariate data is to make informed predictions about the unknown. If a meteorologist models atmospheric warming, a doctor tracks child growth percentiles, or a retail executive forecasts holiday inventory demands, they use trend line equations to predict future or unmeasured values. However, not all predictions carry the same degree of statistical reliability. In this chapter, you will master using linear models to make quantitative predictions, distinguish between safe interpolation and risky extrapolation, and understand the limits of linear models.
Predicting Using a Linear Model
To make a prediction using a linear model equation y = mx + b:
Case 1: Predict y from a given x (Direct Evaluation)
Substitute the given value of x into the equation and compute y.
Example: Given y = 3.5x + 12, predict y when x = 8:
y = 3.5*(8) + 12 = 28 + 12 = 40.
Case 2: Predict x from a given y (Equation Solving)
Substitute the given value of y into the equation and solve for x using inverse operations.
Example: Given y = 3.5x + 12, predict x when y = 47:
47 = 3.5x + 12 ---> 35 = 3.5x ---> x = 10.
The mathematical computation is straightforward; the critical statistical challenge is assessing whether the prediction is an interpolation or an extrapolation.
Interpolation: Predicting Inside the Observed Data Range
Interpolation is the process of estimating a value that lies BETWEEN the minimum and maximum observed values of the independent variable in the original dataset.
Definition of Interpolation:
x_min <= x_prediction <= x_max
Visualizing Interpolation:
Observed Data Domain: x ranges from 5 to 25.
Predicting at x = 14:
x = 14 (INTERPOLATION: Safe and Highly Reliable!)
|
<----+-------------*-----------------------+----> x
x_min = 5 x_max = 25
Because you have actual empirical data points surrounding both sides of your prediction point, you know that the linear relationship holds true across that entire interval. Interpolation carries high statistical confidence and reliability.
Extrapolation: Predicting Outside the Observed Data Range
Extrapolation is the process of estimating a value that lies OUTSIDE the range of the observed data, predicting beyond the known minimum or maximum values:
Definition of Extrapolation:
x_prediction < x_min OR x_prediction > x_max
Visualizing Extrapolation:
Observed Data Domain: x ranges from 5 to 25.
Predicting at x = 45:
x = 45 (EXTRAPOLATION: High Risk!)
|
<----+-------------------------------------+---------*----> x
x_min = 5 x_max = 25
Extrapolation is inherently dangerous and unreliable because it assumes that the linear trend will continue indefinitely into the future or into extreme conditions without changing. In the real world, physical systems almost always hit boundaries: plants reach maximum biological heights, chemical reactions run out of reagents, and economic growth slows down.
Classic Extrapolation Failure:
A puppy weighs 2 pounds at birth, 8 pounds at 2 months, and 14 pounds at 4 months.
Linear Model: Weight = 3 * (Age in Months) + 2.
- Interpolation at 3 months: Weight = 3*(3) + 2 = 11 pounds. (Accurate!)
- Extrapolation at 120 months (10 years): Weight = 3*(120) + 2 = 362 pounds!
A 362-pound dog is biologically impossible; growth levels off after adolescence!
The Domain of Validity
Every empirical linear model has a domain of validity: the specific range of input values over which the model can be reasonably expected to apply. Outside this domain, linear models inevitably break down into nonlinear plateaus or catastrophic limits.
Chapter Exercises
Exercise 1: An experiment tracks the temperature of cooling water in a beaker. The observed data spans elapsed times from t = 2 minutes to t = 18 minutes. The derived linear model is T = -2.5t + 85. Predict the water temperature at t = 10 minutes and at t = 40 minutes. Classify each prediction as interpolation or extrapolation, and evaluate the physical realism of the second prediction.
Exercise 2: A student collects data on running distance (between 1 and 6 miles) and calories burned, deriving the model C = 115d + 20. The student uses the model to predict calories burned for a 26-mile marathon run. Classify this prediction as interpolation or extrapolation and explain why caution is required.
Exercise 3: Explain in paragraph form why interpolation provides vastly higher statistical confidence than extrapolation, discussing underlying trend stability and real-world system boundaries.
Solutions and Explanations
Solution 1: For the cooling water model T = -2.5t + 85, where observed data ranges from t = 2 to t = 18 minutes: To predict temperature at t = 10 minutes, substitute 10 for t: T = -2.5(10) + 85 = -25 + 85 = 60 degrees. Because t = 10 falls between the minimum observed time of 2 minutes and the maximum observed time of 18 minutes (2 <= 10 <= 18), this prediction is classified as an Interpolation. It carries high statistical reliability. To predict temperature at t = 40 minutes, substitute 40 for t: T = -2.5(40) + 85 = -100 + 85 = -15 degrees. Because t = 40 lies far beyond the maximum observed time of 18 minutes (40 > 18), this prediction is classified as an Extrapolation. Regarding physical realism, this prediction is completely impossible: liquid water sitting in a normal room will never cool down to -15 degrees (below freezing); cooling decelerates asymptotically as the water temperature approaches ambient room temperature (approximately 20 degrees Celsius or 68 degrees Fahrenheit). The linear model breaks down completely.
Solution 2: In the calorie expenditure model C = 115d + 20, the observed data domain spans distances from d = 1 mile to d = 6 miles. Predicting the calories burned for a 26-mile marathon run requires evaluating the model at d = 26 miles. Because 26 miles is more than four times greater than the maximum observed distance of 6 miles (26 >> 6), this prediction is an extreme Extrapolation. Extreme caution is required because human physical endurance does not scale indefinitely as a simple linear function. Over the course of a 26-mile marathon, an athlete experiences severe glycogen depletion, dehydration, muscle fatigue, and changes in biomechanical running efficiency. These physiological stress factors cause the true rate of calorie consumption to deviate from a basic short-distance training run, meaning the simple linear model may significantly miscalculate actual energy expenditure.
Solution 3: Interpolation provides vastly higher statistical confidence than extrapolation because it operates entirely within an established, verified empirical envelope. When you interpolate, you possess actual observed data points flanking both sides of your target point, proving conclusively that the linear trend remained stable, continuous, and unbroken across that interval. In contrast, extrapolation ventures into completely uncharted territory where no experimental observations exist. Extrapolation relies on the dangerous assumption that the exact same linear rate of change will continue indefinitely without interference. In the real world, virtually all physical, economic, and biological systems eventually encounter natural constraints and capacity boundaries—such as carrying capacities, physiological limits, saturation points, and friction—that bend straight linear relationships into flat plateaus or steep declines. Therefore, making predictions beyond the observed boundary carries high risk and unquantified uncertainty.