Evidence for Multiple Labor Market Segments: An Entropic Analysis of US Earned Income, 1996-2007

Abstract

This paper revisits the fitting of parametric distributions to earned income data. In line with Camilo Dagum's dictum that candidate distribution should not only be chosen for fit, but that economic content should also play a role, a new candidate is proposed. The fit of a simple finite mixture performs as well or better than the recently widely used generalized beta of the second kind (GB2) and is argued to be easier to interpret economically. Specifically, the good fit is taken as evidence for a finite level of segmentation in the labor market, with a distinctly different generating mechanism underlying each segment. It is speculated that this could be reconciled with either modern labor market models in which agent or firm heterogeneity can lead to different equilibrium configurations, or an older theory of labor market segmentation. In fact, the use of the mixture model addresses one of the central weaknesses of testing that older theory empirically. The approach taken in this paper is also motivated by the work of E. T. Jaynes, the father of maximum entropy approaches to statistical inference and related to the recent work by physicists on the distribution of income.