R²

R-squared (Coefficient of Determination)

The proportion of variance in the outcome explained by a regression model — ranges from 0 (no explanatory power) to 1 (perfect fit).

When you'd see it: Linear regression model evaluation reports. 'The model has an R² of 0.68' means 68% of the variation in the outcome is explained by the model's predictors, leaving 32% unexplained.

Why it matters: R² tells you how much the model's predictors account for the outcome's variation. Context matters: an R² of 0.15 may be meaningful for noisy behavioral data, while an R² of 0.95 should trigger suspicion about data leakage or a predictor that's a near-duplicate of the outcome.

Common mistakes: Treating R² as a standalone measure of model quality. R² always increases (or stays the same) when you add more predictors, even irrelevant ones — which is why adjusted R² exists. Also: high R² in training data but not in holdout data is a sign of overfitting.

Related linear regression, RMSE, overfitting, variance