Predicting with Models: Interpolation vs. Extrapolation
Predicting with models — interpolation versus extrapolation — is taught in Grade 9 Algebra 1, California Reveal Math (Unit 10: Quadratic Functions). Interpolation means predicting within the original data's domain and is highly reliable. Extrapolation predicts outside the domain and carries increasing uncertainty because the trend may not continue. For a population model built on 2010–2020 data, predicting 2015 is reliable interpolation, but predicting 2060 is risky extrapolation that assumes no change in real-world conditions.
Key Concepts
Property Once a best fit regression model is established, you can substitute an input value into the equation to predict a future or missing output. Interpolation: Making a prediction within the domain (range of $x$ values) of the original data. This is highly reliable. Extrapolation: Making a prediction outside the domain of the original data. This carries increasing uncertainty because we must assume the trend continues unchanged indefinitely.
Examples Interpolation (Reliable): An exponential model for a town's population is built using data from the years 2010 to 2020 ($0 \leq t \leq 10$). Using the model to predict the population in 2015 ($t = 5$) is interpolation. It is highly reliable because it falls within the studied timeframe. Extrapolation (Risky): Using that same model to predict the population in the year 2060 ($t = 50$) is extrapolation. It is mathematically possible to calculate, but highly uncertain in reality, as resources, economy, or space might limit population growth long before 2060, changing the curve from exponential to something else. Contextual Failure: A linear regression for monthly toy sales gives $S(m) = 3.2m + 10$ for months 1 through 12. Extrapolating to $m = 3$ produces a mathematical output, but it is contextually impossible because "negative 3 months" has no meaning in this data set.
Explanation A mathematical model is essentially a pattern recognition machine. It is incredibly accurate at filling in the blanks between the data points you already gave it (interpolation). However, asking the model to predict what happens 50 years into the future (extrapolation) is risky. The math equation will blindly continue the curve forever, but in the real world, circumstances change. Always check your extrapolated predictions to see if they actually make logical sense in the context of the story!
Common Questions
What is the difference between interpolation and extrapolation?
Interpolation predicts within the original data range (reliable). Extrapolation predicts outside it (increasingly uncertain). A model built on data from 2010 to 2020 reliably interpolates 2015 but only estimates 2060.
Why is extrapolation risky?
The model assumes the trend continues unchanged indefinitely, but real-world conditions like resources, economy, or policies can change the pattern. The math still produces a number, but that number may be unrealistic.
When does interpolation fail to be reliable?
Interpolation can fail if the data itself is poor quality, if the chosen model type does not fit the data, or if the interpolated point is between data with a hidden local anomaly.
Can extrapolation ever produce contextually impossible results?
Yes. A linear sales model S(m) = 3.2m + 10 extrapolated to m = -3 gives a mathematical output, but negative months have no meaning. Always check extrapolated results against real-world logic.
How do you know which type of prediction a problem is asking for?
Check if the x-value falls inside or outside the range of the original dataset. Inside the range = interpolation. Outside the range = extrapolation. State which type and note the associated reliability.