Nested vs Non-Nested Models
You'll learn how to recognize whether two models are nested or non-nested — a key distinction when choosing the right comparison test.
What Are Nested Models?
Two models are nested when one model is a special case of the other. In other words, you can create the simpler model by removing (or setting to zero) one or more predictors from the more complex model.
Example:
- Model A:
price = β₀ + β₁·size + β₂·bedrooms + β₃·age - Model B:
price = β₀ + β₁·size + β₂·bedrooms
Model B is nested within Model A because you get Model B by dropping the age predictor from Model A. Model A is the "full model" and Model B is the "reduced model."
What Are Non-Nested Models?
Two models are non-nested when neither is a special case of the other. They use different predictors that don't have a subset relationship.
Example:
- Model X:
price = β₀ + β₁·size + β₂·bedrooms - Model Y:
price = β₀ + β₁·location + β₂·age
Neither model can be obtained by simply dropping variables from the other — they represent fundamentally different modeling approaches.
Why Does This Matter?
The distinction determines which comparison tools you can use:
- For nested models: Use the Partial F-Test (which you learned in lesson 623) to formally test whether the extra predictors significantly improve the model
- For non-nested models: Compare using Adjusted R-Squared, AIC, or BIC — but you cannot use the Partial F-Test
Think of it like this: nested models are like Russian dolls (one fits inside another), while non-nested models are like different puzzles that can't be compared piece-by-piece.
Key Takeaway: Nested models have a subset relationship where one is a simpler version of the other; non-nested models use different predictors and require different comparison methods.