A new machine learning-based treatment bite for long run minimum wage evaluations

Empirical evaluations of national minimum wages, such as in Germany or the UK, rely on bite measures that capture treatment variation; measured from the incidence (or intensity) of employees paid below the threshold before the minimum wage was introduced or raised. Bite-dependent estimations face the problem of dynamic selection, implying that even in the absence of the minimum wage the bite may have changed over time. We apply a machine learning method from the field of regularized regression to predict the contemporary bite of the German minimum wage, allowing us to address unobserved dynamic selection in an empirical evaluation of long run effects of the minimum wage. Our LASSO predicted bites show clear improvements over simple forward updating of the initial bite, allowing us to estimate contemporary effects of the German minimum wage from 2015 to 2017.