R-squared (Coefficient of Determination)

R-squared is a 0-to-1 score (often read as a percentage) for how much of the variation in the outcome your model accounts for. An R-squared of 0.70 means the model explains 70% of the ups and downs in y, leaving 30% to everything it does not capture.

It is handy for a quick sense of fit, but it is widely abused. A high R-squared does not mean the model is correct, useful for prediction, or causal; and adding more predictors can only push R-squared up, even useless ones, which is why analysts watch adjusted R-squared (which penalizes extra variables) and test on held-out data instead of trusting in-sample R-squared alone.