Interpreting Adjusted R-Squared Values
You'll learn what adjusted R-squared actually tells you about your model's quality and when to trust it over regular R-squared.
The Core Idea
Adjusted R-squared gives you a penalized measure of model fit that accounts for the number of predictors you've added. Unlike regular R-squared (which can only stay the same or increase when you add variables), adjusted R-squared can actually decrease if you add a predictor that doesn't pull its weight.
Think of it like this: Regular R-squared is an overly generous grader who gives you credit just for showing up with more variables. Adjusted R-squared is a fair grader who asks, "Does this new variable actually improve understanding, or are you just padding your model?"
What the Values Mean
- Higher values (closer to 1.0) indicate your predictors explain more variance in the outcome, accounting for model complexity
- Lower values (closer to 0) suggest weak explanatory power, even after considering how many predictors you included
- Decreasing adjusted R-squared when you add a variable means that variable adds more noise than signal—it's not worth including
When It Can Decrease
Adjusted R-squared will drop when you add a predictor whose contribution to explaining variance is too small to justify the added complexity. This is its superpower: it stops you from mindlessly stuffing your model with irrelevant variables.
Key Takeaway: Adjusted R-squared helps you balance model fit against complexity—use it to decide whether adding another predictor truly improves your model or just inflates it with unnecessary variables.