Modeling Income Distributions Using Elevated Distributions

We present a two parameter family of continuous distribution on a bounded domain which has an elevated but finite density value at its lower bound. Such a characteristic appears to be useful, for example, when representing income distributions at lower income ranges. The family generalizes the one parameter Topp and Leone distribution originated in the 1950's and recently rediscovered. The family of beta distributions has been used for modeling bounded income distribution phenomena, but it only allows for an infinite and zero density values at its lower bound, and a constant density in case of its uniform member. The proposed family alleviates this apparent jump discontinuity at the lower bound. The U.S. Income distribution data for the year 2001 is used to fit distributions for Caucasian (Non-Hispanic), Hispanic and African-American populations via a maximum likelihood procedure. The results reveal stochastic ordering when comparing the Caucasian (Non-Hispanic) income distribution to that of the Hispanic or African-American population. The latter indicates that although substantial advances have reportedly been made in reducing the income distribution gap amongst different ethnic groups in the U.S. during the last 20 years or so, these differences still exist.