The Distribution of Wages: A Non-parametric Decomposition

This paper presents a non-parametric procedure to analyze the effects of different factors on observed movements in any distribution. These effects are estimated by applying kernel density methods to weighted samples in order to obtain counterfactual distributions. The advantage of this approach is that it provides a direct means of investigating if these factors have an impact and where in the density they do so, and it offers a new decomposition method of within and between group components. The approach to the decomposition analysis applied in this paper differs from the classical one of additively decomposable inequality indexes. If the purpose of the analysis is to understand what determined the variation in relative inequality, there is no doubt that the decomposition of the indexes belonging to the generalized entropy family is the best method. If, instead, the aim is to monitor what factors modified the entire distribution, where precisely on the distribution these factors had an effect, and what determined the variation in the level of polarization observed, then that method is useless. The non-parametric method proposed is the one to use, but with one caveat: All the results assume that there are no general equilibrium effects. The paper contains summary statistics of the observed movements and of distance and divergence among the estimated and counterfactual distributions; an original modification of an index of polarization; and an application of the method to the Italian distribution of wages.