Adjusted R Square

Adding a predictor can raise ordinary R Square because the expanded model explains at least as much variation as before. Adjusted R Square also considers whether the added variable provides enough useful information to justify the model’s increased complexity. When its contribution is small relative to that complexity, the adjusted measure decreases, signaling that the expansion may not improve the model meaningfully.

Adjusted R Square incorporates both the sample size and the number of explanatory variables when evaluating explained variation. Consequently, the same apparent improvement in model fit can receive different adjustments in datasets with different sample sizes or model sizes. This dependence makes the measure more informative than unadjusted R Square when assessing whether additional predictors provide sufficient explanatory value.

The penalty represents a correction for model complexity rather than a judgment that every additional variable is invalid. A predictor must contribute useful explanatory information to offset the adjustment associated with expanding the model. This feature discourages treating a small increase in explained variation as evidence of a better model when that increase comes from adding unnecessary explanatory variables.

Researchers should compare Adjusted R Square when candidate regression models contain different numbers of predictors. Ordinary R Square can increase simply because variables were added, whereas the adjusted measure accounts for that expansion. Examining the adjusted values helps identify the model that offers a more balanced combination of explanatory power and parsimony, rather than favoring the largest model automatically.

A practical comparison begins by fitting the competing regression models and recording their Adjusted R Square values along with their predictor counts. Researchers then assess whether a larger model produces a meaningful improvement after the complexity adjustment. The comparison supports selecting a model that explains variation efficiently, especially when one candidate includes additional variables whose contribution may be limited.

Adjusted R Square helps address overfitting by making model complexity part of the evaluation. A model may appear stronger under ordinary R Square after accumulating predictors, even when those variables add little useful information. If the adjusted value falls or fails to improve, researchers receive evidence to question whether the added complexity improves the model’s explanatory balance.