On the one hand, it has been well established over the last decade that the worldwide income distribution has been polarizing or stratifying into distinct classes of income since World War II (see, among others, Bianchi (1997), Desdoigts (1994), and Quah (1996)). The analysis of the world income distribution is motivated by the following key question: can we find in the data evidences of poverty traps? Such a question can be found for instance in Baumol (1986)'s idea of convergence clubs.
On the other hand, an extensive empirical literature investigates the sources of growth and convergence through the estimation of worldwide cross country growth regressions using explicitly formulated growth models (see, among others, Mankiw, Romer, and Weil (1992), and Durlauf and Quah (1998) for an extensive and insightful survey on ``why do growth rates differ?'').
As Quah (1996) stresses, growth regressions average across the cross-section but they can only give a picture of the behavior of the conditional mean, not of the whole distribution. This article illustrates how traditional cross-country growth regressions can be used to analyze the immediate sources of the worldwide income distribution dynamics over the period 1960-1985. As XploRe provides an extensive set of parametric and nonparametric methods, it is a natural statistical computing environment to investigate international growth differences and changes in the world income distribution.
The analysis starts with an estimation of a
classical cross-country growth regression such as found by Temple (1998)
who investigates the correlation between equipment investment and economic
growth, and its compatibility with the Solow (1956) growth model from which
an explicit convergence equation is derived. The estimated model and the
underlying data are then used to quantify the immediate sources of each
country's differential growth performance. In a second step, a nonparametric
counterfactual exercise is proposed that allows us to analyze the
effects of the various explanatory variables on changes in the world income
distribution. The effects of the different variables are estimated by
applying kernel density methods. The procedure provides a visually clear
representation of where in the density of incomes the specified factors
exert the greatest impact. (See also Di Nardo, Fortin, and Lemieux (1996) for an analysis of
the effects of institutional and labor market factors on the
US distribution of wages, 1973-1992.)