Use of regression trees in the study of nonparametric wage structures
This study is concerned with the application of multivariate nonparametric models known as regression trees to the analysis of the U.S. wage structure. In Chapter 1, I first review regression trees and other available multivariate nonparametric techniques, highlighting their differences and common features. In the second part of Chapter 1, I look at the literature on the U.S. wage structure in connection with the issue of functional specification and argue that regression trees is particularly well suited for analyzing wage structures. In Chapter 2, I implement regression trees on U.S. wages for white male workers to estimate experience-wage profiles and unveil local sudden breaks in the profiles at the end of the working life. For 1980, these breaks account for about 50% of the negative average differential between the last two experience groups. This effect decreases continuously until 1995. In Chapter 3 I propose a simple extension of the Oaxaca-type average wage gap decompositions between any two groups of workers. This procedure can be carried out without any compromise in the interpretation using a nonparametric wage structure. I then study wage gap decompositions for Mexican workers in the U.S. labor market. Finally, in Chapter 4 I apply regression trees to study both the relative growth performance of workers' real wages and the sources of wage dispersion and its evolution in the U.S. from 1980 onwards. On trends, the technique uncovers a linear structure for the growth experience of white workers with less than forty years of experience. On dispersion, at least 10% of the increase in observed variance came from changes in the structure of wages itself.