Along with Phil Magness, John Moore and Phil Vøn Schløsser, I assembled a series of concerns that we have regarding the measurement of income inequality in the US before 1945 as pictured by the Piketty-Saez U-Curve. We argue that the pronounced left-hand side of the U-Curve is sensitive to minor changes in assumptions as well as minor improvements in data quality. We argue that income inequality probably did fall and rise over the 20th century, but not in the proportions presented in the Piketty-Saez papers. The paper is available here on SSRN and the abstract is below:
In this article, we reconsider the level and trend of the income inequality series produced by Piketty and Saez (2003, 2015) for the United States using tax data for the period prior to 1945, which forms the left-side of a century-long distributional U-curve. We argue that there are reasons to doubt the depicted shape of the left-side of the U-curve of inequality. We make corrections to the series for reporting behavior conditional on tax regimes, minor changes to the definition of the tax unit population, and the fiscal income denominator. All of these corrections show comparatively stable top income shares throughout the period. We point out that comparisons with other state-level tax datasets – like Wisconsin which has conceptual advantages over the IRS data – yield dramatically different results that should make us skeptical of the trends observed with the IRS data.