7.1. Summary

The main contribution of this dissertation is the construction of a new homogeneous set of panel data by state cross-sections and annually from 1913 to 2003, using the Statistics of Income publications by the U.S. Internal Revenue Service. This database represents well the top 10 percent of the income distribution, but data from other sources are needed to account for average income. Meanwhile, the top decile database offers an alternative way to average income figures used by Barro and Sala-i-Martin to study the same topic.

In order to address the issue of income convergence across the United States over the long-run, three types of convergence are distinguished: 1) the β convergence of average income in comparison with the β convergence of income in the top decile (growth and inequality regressions), 2) the σ convergence (dispersion of average and top incomes), and 3) the convergence of top incomes towards the lower fractiles of the income distribution.

In the case of the β convergence, we found evidence confirming conclusions towards convergence within the top decile, and more mitigated results for convergence among state average incomes. More particularly, the β convergence based on average income does not seem to hold in recessionary phases of the Juglar cycle (fully occurring within a period of about ten years). To compare growth regressions to inequality regressions, where inequality indicators are based on the top income series, the results showed that income inequality is positively correlated to the average income, and negatively correlated to economic growth rates.

In the case of the σ convergence, the trend over time opposes two groups of income. On the one hand, average income and income of the top percentile both recorded a decline in dispersion across states (except after the mid 1980s). On the other hand, the incomes of fractiles 90-95, and 95-99 percent were featured with a rise in dispersion across states.

Finally, the convergence (or divergence) of the top decile towards (or away from) the bottom decile is emphasized from 1965 to 2003. Similarly, the top and bottom quartiles are compared as well. The dispersion indicators of the lower layers of the income distribution were estimated by extending the Pareto assumption from the top decile to the full income distribution. What emerges from these estimates is that the top income shares did not grew faster than the low income shares from 1965 to 1984, then this trend got totally inversed and reached a peak in 1988 in all states, they decreased again until 1985, and finally reached a local maximum in 2000.