Comparison of personal income inequality estimates based on data from the IRS and Census Bureau

This paper demonstrates quantitatively that modern estimates of income inequality based on the data reported by the IRS are not reliable. Principal problem of the IRS data consists in highly volatile income estimates in the low-end of personal income distribution. This volatility is likely related to measurement errors, changes in definitions or improper reporting. Personal income estimates at high and the highest incomes are robust and follow the Pareto law. At high incomes, personal income distributions for 1990 and 2004, when normalized to total population with income and total (gross) personal income, practically coincide. Hence, the inequality estimates based on the IRS data are distorted by inaccurate readings in the low-income zone. At the same time, income data provided by the US Census Bureau are consistent over time in all income ranges. Results presented by Kitov (2007) demonstrate that personal income distributions based on readings obtained in the Current Population Survey are characterized by practically constant Gini coefficient since 1960. This observation implies that normalized personal income distributions are also not changing with time.