We examine the dynamic evolution of incomes, both disposable and gross, for several groups in the PSID panel data at several points from 1968 to 1997. We employ the extended Kolmogorov-Smirnov tests of First and Second Order Stochastic Dominance (SD) as implemented by Maasoumi and Heshmati (2000). They do not impose the Least Favorable Case (LFC) of the composite null hypotheses of SD orders. This is in contrast to simulation and bootstrap-based techniques that do so, resulting in tests that are not asymptotically similar or unbiased. Our approach is also different from the subsampling technique of Linton et al (2005) who obtain critical values for these tests under very general sampling schemes. We offer partial control for many individual/family specific attributes, such as age, gender, education, number of children, work and marital status, by comparing group cells. This avoids having to specify and estimate models of dependence of incomes on these attributes, but lacks the multiple controls that is the promise of such techniques. We find a surprising number of strong rankings, both between groups and over time, in gross income and, to a lesser extent, in 'disposable' incomes.
Stochastic Dominance bootstrap income distribution testing PSID gender education age marital status