The Dodgers' rookie tossed each player's jersey number into a runs model and declared victory.
0 / 2 graded
person
Rookie's submission
Comparing two models of runs
Model A (HR + walks): R² = 0.92
Model B (HR + walks + jersey#): R² = 0.94
"Adding jersey number improved the model!"
1 · Why is “R² went up, so it's better? wrong here?
Every extra column gives the fit more freedom, so R² never goes down. That's exactly why it can't tell you whether a new predictor actually helps.
2 · What should the rookie check instead?
Adjusted R² only rises when a predictor earns its place. A nonsense variable like jersey number makes it drop — the honest signal that Model B is worse.
trending_down
Graded — R² always climbs.
Adding predictors can't lower R², so use adjusted R² (or a test on the new term) to judge whether one belongs.
