R-squared

Also known as · R squared

The share of the variation in yy that the regression explains, R2=1−RSS/TSSR^2 = 1 - RSS/TSS (equivalently MSS/TSSMSS/TSS). It never falls when a regressor is added, even a noise one, and it depends on the variance of yy, so it only compares models with the same outcome variable.

When to use

Quote it to describe fit, and switch to Adjusted R-squared when comparing models with different numbers of regressors. The overall model F-test can also be written purely in terms of R2R^2.

See Lecture 8.

Appears in