Course contentsShow
← Data Science
Lesson 684 of 2,14518. Generalized Linear ModelsFree lesson

Likelihood Ratio Tests for Model Comparison

Comparing nested logistic regression models using the likelihood ratio test to assess predictor importance.

Likelihood Ratio Tests for Model Comparison

You'll learn how to use likelihood ratio tests to formally compare nested logistic regression models and decide whether additional predictors improve your model.

The Core Idea

When you have two logistic regression models—one simpler (fewer predictors) and one more complex (additional predictors)—you need a rigorous way to decide if the extra complexity is justified. The likelihood ratio test (LRT) compares how well each model fits the data by examining their likelihoods.

Think of it like comparing two explanations for a mystery: one simple, one detailed. The LRT asks, "Does the detailed explanation fit the facts significantly better than the simple one, or is the extra detail just noise?"

How It Works

The test compares the maximum likelihood of the restricted model (fewer predictors) against the full model (more predictors). The test statistic is:

LR = -2 × [log(L_restricted) - log(L_full)] = -2 × log(L_restricted / L_full)

This statistic follows a chi-squared distribution with degrees of freedom equal to the difference in the number of parameters between models.

  • Large LR value → Full model fits much better → Reject the simpler model
  • Small LR value → Models fit similarly → Keep the simpler model (parsimony wins)

Practical Use

If you're deciding whether adding age and income to a model that already includes education improves prediction of loan default, the LRT gives you a p-value. A small p-value (typically < 0.05) means the extra predictors significantly improve the model.

Key Takeaway: The likelihood ratio test provides a formal statistical framework for comparing nested logistic regression models, helping you determine whether additional predictors meaningfully improve model fit beyond chance.