Abstract

Most empirical evaluations of national minimum wages rely on a bite measure that captures treatment variation. Bite-dependent estimates face the problem of dynamic selection. That is, the treatment bite can change over time even in absence of a minimum wage. We apply machine learning methods to predict the contemporary bite of the German minimum wage, thereby accounting for unobserved dynamic selection. In an empirical evaluation, LASSO-predicted bite measures show significant improvements over conventional time-constant measures. When estimating contemporary wage effects of the German minimum wage introduction, wage increases are positive but smaller compared with conventional estimation.