Introduction to Competing Function Model Validation

In our previous lessons, we learned how to build linear and exponential models. But what happens when a set of data looks like it could be modeled by either a line or a curve? How do we decide which one is "correct"? In AP Precalculus, we call this process Competing Function Model Validation. It is the art of using data and mathematical tools to justify why one function fits better than another. Don't worry if this seems a bit subjective at first; there are specific tools we use to make the best choice!

1. Understanding the Competition

When we look at a scatterplot, multiple function types might seem to "fit" the general trend. In Unit 2, we primarily compare linear models and exponential models, but we also consider quadratic, cubic, or logarithmic regressions as competitors.

To differentiate between them, we look at the rate of change:

  • Linear Models: These are best if the data shows a constant rate of change (the outputs increase/decrease by the same amount for every equal step in input).
  • Exponential Models: These are best if the data shows a proportional rate of change (the outputs are multiplied by a constant ratio for every equal step in input).

Quick Review: If you need a refresher on the differences between these growth patterns, check out Chapter 2.2: Change in Linear and Exponential Functions.

2. The Most Powerful Tool: Residuals

The most important way to validate a model in this course is by looking at residuals. A residual is simply the vertical distance between the actual data point and the value predicted by your model.

\( \text{Residual} = \text{Actual } y\text{-value} - \text{Predicted } y\text{-value} \)

\( \text{Residual} = y - \hat{y} \)

How to Read a Residual Plot

A residual plot graphs the residuals on the vertical axis and the input \( x \) on the horizontal axis. This plot tells us if our chosen model is appropriate:

  • Good Fit: If the points in the residual plot are randomly scattered around the horizontal axis (zero line) with no clear pattern, the model is likely a good fit.
  • Poor Fit: If the points in the residual plot form a clear pattern (like a "U" shape or a curve), it means the model is systematically missing something. Even if the points look "close" to the line in the original scatterplot, a pattern in the residuals suggests a different type of function would be better.

Memory Aid: Think of residuals like "errors." If your errors are random, you're just experiencing natural noise. If your errors have a pattern, you've made a systematic mistake in picking your function type!

3. Using Technology for Regression

For the AP Exam, you are required to use a graphing calculator (in Section I Part B and Section II Part A) to find these models. The calculator uses a method called "Least Squares Regression" to find the line or curve of best fit.

Steps for Model Validation using Technology:
  1. Enter Data: Put your \( x \) and \( y \) values into your calculator lists.
  2. Calculate Regressions: Run a Linear Regression \( (y = ax + b) \) and an Exponential Regression \( (y = a \cdot b^x) \).
  3. Plot Residuals: Most calculators can automatically generate a list of residuals for each model. Create a scatterplot of these residuals.
  4. Compare: Look for the model whose residuals are most randomly distributed.

Did you know? Even if a model has a very high "correlation" or looks like it touches almost all the points, if the residuals show a curve, a different model is technically better for making long-term predictions.

4. Comparing Models in Context

Sometimes, math isn't the only factor—the context of the problem matters too. When validating a model, ask yourself:

  • End Behavior: Does the function make sense as \( x \) gets very large? A linear model might predict a population becomes negative (impossible), while an exponential model might predict it grows to infinity.
  • Initial Values: Does the \( y \)-intercept (\( x = 0 \)) make sense for the scenario?
  • Domain Restrictions: Are there physical limits (like a maximum capacity) that make one model more realistic than another?

Example: If you are modeling the cooling of a cup of coffee, a linear model would suggest the coffee eventually reaches \( -500 \) degrees, which is impossible. An exponential model (specifically one that levels off at room temperature) is much more realistic.

5. Common Mistakes to Avoid

Mistake 1: Relying only on the scatterplot. Some exponential curves look very flat and can be mistaken for lines if you don't look at the residuals.

Mistake 2: Mixing up "Actual" and "Predicted." Always subtract the model's value (\( \hat{y} \)) from the data's value (\( y \)). If the residual is positive, the data point is above the model. If it's negative, the data point is below the model.

Mistake 3: Forgetting Units. When explaining why a model is valid in a Free Response Question (FRQ), always refer back to the units of the problem (e.g., "The model predicts the height in meters...").

Summary Key Takeaways

  • Validation is the process of choosing the best model among "competitors" like linear and exponential functions.
  • Residuals are the key evidence. \( \text{Actual} - \text{Predicted} = \text{Residual} \).
  • Random residuals = Good model. Patterned residuals = Bad model (pick a different function type).
  • Technology is used to calculate regression equations and view residual plots.
  • Context (real-world logic) must support your mathematical choice.