Fig. 1.

A simulation example with $N=100$ observations and $p=10$ or 50 predictors and null effects ($\beta =0$). Shown are the nominal and actual P values for entering the strongest predictor at the first step; the 45° dashed line is included for reference.