CFDS Discussion Paper 2026/3
Abstract
This paper documents systematic and economically meaningful differences across the two most widely used cross-country inequality databases, SWIID and WID. Using the full set of overlapping country-year observations since 1980, we show that agreement between the databases is limited not only in levels but, more importantly, in within-country dynamics. For market-income inequality in particular, correlations decline sharply over time and are close to zero in recent decades. We further demonstrate that established empirical findings are highly sensitive to the choice of inequality data. The divergence between databases cannot be reduced to simple definitional differences such as population coverage or equivalization. Taken together, our results imply that database choice is not a second-order technical detail but a central determinant of empirical conclusions in the inequality literature.