Comparisons of penalized least squares methods by simulations

摘要

Penalized least squares methods are commonly used for simultaneous estimationand variable selection in high-dimensional linear models. In this paper wecompare several prevailing methods including the lasso, nonnegative garrote,and SCAD in this area through Monte Carlo simulations. Criterion for evaluatingthese methods in terms of variable selection and estimation are presented. Thispaper focuses on the traditional n > p cases. For larger p, our results arestill helpful to practitioners after the dimensionality is reduced by ascreening method. K