Robust inference on income inequality: t-statistic based approach

Empirical analyses of income and wealth inequality often face the difficulty that the observations are heterogeneous, heavy-tailed, or correlated in some unknown fashion. This article focuses on applications of the recently developed computationally simple t-statistic based robust inference approach to the analysis of inequality. Two regions can be compared in terms of inequality as follows: the data in the samples relating to the two regions are partitioned into small numbers of groups, and the chosen inequality index/measure is estimated for each group. Inference is then based on standard t-tests with the resulting group estimators. The t-statistic based approach results in valid inference, as long as the group estimators of the inequality index are asymptotically independent, unbiased, and Gaussian, possibly with different variances. These conditions are typically satisfied in empirical applications. The presented method complements and compares favorably with other approaches to inference on inequality. We apply this approach to examine income inequality across Russian regions. Our analysis reveals that income distribution in Russia is notably heavy-tailed, with most regions exhibiting higher levels of inequality compared to Moscow. Robust comparisons of this type offer a good foundation for evaluating and shaping regional policies aimed at addressing income disparities.