Anova: Predicted R-Squared¶

Predicted R-squared is a measure of how well the model predicts a response value. It is computed as

$Pred.\,R^2=1-\left [ \frac{PRESS}{SS_{residual}+SS_{model}} \right ]=1-\left [ \frac{PRESS}{SS_{total}-SS_{curvature}-SS_{block}} \right ]$

The Adjusted R-squared and Predicted R-squared should be within approximately 0.20 of each other to be in “reasonable agreement.” If they are not, there may be a problem with either the data or the model.

“Anova: PRESS” on page 1