Score: 0

Bias in Gini coefficient estimation for gamma mixture populations

Published: March 2, 2025 | arXiv ID: 2503.00690v3

By: Roberto Vila, Helton Saulo

Potential Business Impact:

Fixes unfairness measurement for mixed groups.

Business Areas:
A/B Testing Data and Analytics

This paper examines the properties of the Gini coefficient estimator for gamma mixture populations and reveals the presence of bias. In contrast, we show that sampling from a gamma distribution yields an unbiased estimator, consistent with prior research (Baydil et al., 2025). We derive an explicit bias expression for the Gini coefficient in gamma mixture populations, which serves as the foundation for proposing a bias-corrected Gini estimator. We conduct a Monte Carlo simulation study to evaluate the behavior of the bias-corrected Gini estimator.

Page Count
15 pages

Category
Statistics:
Methodology