Fig. 1.

Comparison of the true Lasso distribution with the residual bootstrap approximation of the Lasso distribution. The black curve represents the empirical “true” Lasso distribution of $T_1=\sqrt{n}({\widehat{\beta }}_{\lambda ,1}-{\beta }_1)$, over 300 simulated datasets. The blue curve combines the residual bootstrap distribution of $T_1^{\ast }=\sqrt{n}({\widehat{\beta }}_{\lambda ,1}^{\ast }-{\widehat{\beta }}_{\lambda ,1})$ from all 300 datasets