📊 statistics

Unbiased estimation in new Gini index extensions under gamma distributions, with application to real income data

This paper introduces two flexible, position-oriented extensions of the classical Gini index based on order statistics, establishes their theoretical properties including exact unbiasedness for gamma distributions, and demonstrates their superior ability to characterize inequality through simulations and an application to South American GDP data.

Original authors: Roberto Vila, Helton Saulo

Published 2026-06-05

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Roberto Vila, Helton Saulo

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to measure how "uneven" a pile of sand is. For decades, economists have used a single ruler called the Gini Index to do this. It gives you one number: a score of 0 means everyone has the exact same amount of sand, and a score of 1 means one person has it all.

But here's the problem with that single ruler: it's a bit like looking at a mountain range from space. You can see the total height difference, but you can't tell if the steep cliffs are at the bottom (the poor) or the top (the rich). Two different piles of sand could have the same "unevenness" score, even if one has a tiny valley at the bottom and the other has a massive peak at the top.

This paper introduces two new, more flexible rulers to fix that blind spot. The authors, Roberto Vila and Helton Saulo, call them the Extended Lower Gini Index and the Extended Upper Gini Index.

The New "Magnifying Glasses"

Think of the old Gini Index as a wide-angle lens that captures the whole picture but blurs the details. The new indices are like two different magnifying glasses:

The Lower Gini (The "Bottom-Up" Lens): This measures how far a random person is from the poorest person in a small group. It focuses specifically on the "valley" at the bottom of the income distribution. It asks: "How much does the average person have more than the person at the very bottom?"
The Upper Gini (The "Top-Down" Lens): This measures how far a random person is from the richest person in that same group. It focuses on the "peak" at the top. It asks: "How much does the richest person have more than the average person?"

How They Work (The "Group of Friends" Analogy)

To understand how these work, imagine you gather a group of $m$ friends (where $m$ is at least 2) and look at their bank accounts.

The Old Way: You just look at the difference between the richest and poorest friend in the group. That's the standard Gini.
The New Way: You pick one specific friend (let's call them Alex).
- For the Lower Index, you measure the gap between Alex and the poorest person in the group.
- For the Upper Index, you measure the gap between the richest person in the group and Alex.

The paper proves that even though you picked a specific "Alex," the math works out so that the result represents the whole group fairly, no matter who Alex is.

The "Gamma" Secret Sauce

The authors didn't just invent these new rulers; they proved they are mathematically perfect for a specific type of data distribution called the Gamma distribution.

Think of the Gamma distribution as a very common shape that income data often takes (it's skewed, meaning most people are in the middle, but there are a few very rich outliers). The authors showed that for this specific shape of data, their new rulers are unbiased.

In everyday terms, "unbiased" means the ruler doesn't lie. If you use it on a small sample of data, it doesn't systematically overestimate or underestimate the true inequality. It hits the bullseye on average. This is a big deal because many statistical tools only work perfectly if you have a massive amount of data; these new ones work well even with smaller groups.

Testing the Rulers

The authors tested their idea in two ways:

Computer Simulations: They created thousands of fake income datasets on a computer. They checked if their new rulers gave the correct answers. The result? The rulers were accurate, and as they fed them more data, the answers got even more precise.
Real-World Test: They applied these new rulers to the 2023 GDP per capita (income per person) of 11 South American countries.
- The Result: The new indices gave a much richer picture than the old one. They could show that some countries had specific issues at the very bottom of the income ladder, while others had extreme gaps at the very top. The old single number would have just said "inequality is X," hiding these specific details.

The Takeaway

This paper doesn't just say "inequality is bad." It gives economists a new toolkit to measure where the inequality is happening.

If a government wants to know if their policies are helping the poorest, they can look at the Lower Gini.
If they want to know if the wealth gap at the top is widening, they can look at the Upper Gini.

By splitting the measurement into "bottom" and "top," the authors provide a way to see the shape of the income distribution in high definition, rather than just a blurry, single-number snapshot.

Technical Summary: Unbiased Estimation in New Gini Index Extensions under Gamma Distributions

Problem Statement
The classical Gini index, while widely used for measuring income inequality, condenses the entire distribution into a single scalar value. This aggregation can obscure critical distributional features, such as whether inequality is concentrated in the lower, middle, or upper tails. Recent extensions, such as the $m$ -th Gini index (Gavilan-Ruiz et al., 2024), generalize the classical measure by considering the expected range of samples of size $m$ . However, these measures remain aggregate indicators and do not provide insight into how inequality is distributed relative to specific positions within the sample, particularly concerning the lower and upper extremes. There is a need for measures that offer a "position-oriented" assessment of inequality while maintaining desirable statistical properties, such as unbiasedness, under specific distributional assumptions common in economic modeling.

Methodology
The authors introduce two new measures: the extended lower Gini index ( $IG_{m;\min}$ ) and the extended upper Gini index ( $IG_{m;\max}$ ). These are defined for a non-negative random variable $X$ with mean $\mu$ and a sample size $m \geq 2$ :

$IG_{m;\min} = \frac{E[X_1 - \min\{X_1, \dots, X_m\}]}{m\mu}$
$IG_{m;\max} = \frac{E[\max\{X_1, \dots, X_m\} - X_1]}{m\mu}$

These indices quantify inequality relative to the lower and upper tails, respectively. Notably, they sum to the $m$ -th Gini index ( $IG_m$ ). The population coefficients are invariant to the specific observation chosen due to exchangeability, though sample estimators depend on the selected position $i$ .

The paper focuses on Gamma-distributed populations ( $X \sim \text{Gamma}(\alpha, \lambda)$ ), a choice motivated by the distribution's prevalence in modeling income and its analytical tractability regarding order statistics. The authors:

Derive Estimators: Construct sample-based estimators ( $\widehat{d}_i IG_{m;\min}$ and $\widehat{d}_i IG_{m;\max}$ ) using U-statistics over all possible subsamples of size $m$ from a sample of size $n$ .
Establish Asymptotic Properties: Prove consistency and asymptotic normality under mild regularity conditions using the strong law of large numbers for U-statistics and the Hoeffding decomposition.
Prove Unbiasedness: Utilize Laplace transform techniques and specific properties of the Gamma distribution to derive exact expressions for the expectations of the estimators. The authors demonstrate that for Gamma populations, the estimators are unbiased, i.e., $E[\widehat{d}_i IG_{m;\min}] = IG_{m;\min}$ and $E[\widehat{d}_i IG_{m;\max}] = IG_{m;\max}$ .
Characterize Properties: Establish scale invariance, lack of translation invariance, and covariance representations involving the distribution function $F$ .

Key Results

Theoretical Unbiasedness: The paper provides exact finite-sample expressions for the expectations of the proposed estimators under the Gamma model, proving they are unbiased. This extends previous results by Deltas (2003) and Baydil et al. (2025) regarding the $m$ -th Gini index.
Decomposition: The sum of the extended lower and upper indices recovers the $m$ -th Gini index, allowing for a decomposition of overall inequality into lower-tail and upper-tail components.
Simulation Performance: Monte Carlo simulations (500 replications) for Gamma($2, 1$) data with sample sizes $n \in \{10, 30, 50, 100, 200\}$ and $m=3$ confirm the theoretical findings. The estimators exhibit small empirical bias that decreases with sample size and Mean Squared Error (MSE) that converges as expected, supporting the consistency and unbiasedness claims.
Empirical Application: The methodology was applied to 2023 GDP per capita data from 11 South American countries. The data were fitted to a Gamma distribution (validated via KS and Cramér-von Mises tests). Heatmaps of the estimated indices across different $m$ $m$ and $i$ $i$ values revealed systematic patterns:
- The extended lower index increased with $i$ and $m$ , reflecting sensitivity to spread in the lower tail.
- The extended upper index yielded higher values than the lower index for the same parameters, reflecting the greater impact of the maximum order statistic.
- The analysis demonstrated that these indices provide a richer characterization of inequality than a single aggregate coefficient, revealing distinct features of the income distribution at both tails.

Significance and Claims
The authors claim that these are the first Gini-type measures that explicitly separate inequality contributions relative to lower and upper sample extremes while possessing both exact finite-sample unbiasedness under Gamma models and standard large-sample inferential properties.

The paper emphasizes that unlike traditional Gini measures, the proposed indices offer a "position-oriented perspective," allowing researchers to assess how inequality manifests relative to specific distributional tails. By providing unbiased estimators for Gamma populations—a common model for economic data—the authors argue that these measures serve as useful complementary tools for inequality analysis, capable of revealing distributional features that may remain hidden when using a single aggregate coefficient. The work extends the literature on Gini-type measures by bridging the gap between aggregate range-based measures and position-specific inequality assessment.