📊 statistics

Income inequality estimation with gamma mixtures

This paper derives closed-form expressions and establishes the asymptotic properties of an estimator for the $m$ th Gini index under finite mixtures of gamma distributions, validating its performance through Monte Carlo simulations and a real-world income dataset analysis.

Original authors: Roberto Vila, Helton Saulo, Felipe Quintino

Published 2026-07-08

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Roberto Vila, Helton Saulo, Felipe Quintino

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 unevenly a bag of candy is distributed among a group of friends. In economics, this "unevenness" is called income inequality, and the most famous tool for measuring it is the Gini coefficient.

This paper is like a team of statisticians building a better, more flexible ruler to measure that candy distribution, specifically when the "candy" (income) follows a pattern called a Gamma distribution.

Here is a breakdown of what they did, using simple analogies:

1. The Problem: One Size Doesn't Fit All

Real-world income data is tricky. It's not a smooth, perfect bell curve. Instead, it looks like a long tail: most people have a "normal" amount of money, but a few people have massive amounts, stretching the tail far out to the right.

To handle this, the authors use Gamma Mixtures.

The Analogy: Imagine the population isn't one big group, but a mix of three different clubs: a "Low-Income Club," a "Middle-Income Club," and a "High-Income Club." Each club has its own internal rules for how money is distributed. A "Gamma Mixture" is just a mathematical way of saying, "We are looking at a crowd made of several different types of people mixed together."

2. The New Tool: The " $m$ -th" Gini Index

The traditional Gini coefficient compares two people at a time (e.g., "How different is Person A's income from Person B's?").

The Innovation: This paper introduces the $m$ -th Gini Index. Instead of just comparing two people, this new tool grabs a group of $m$ people at once and looks at the difference between the richest and poorest person in that specific group.
Why it matters: It's like checking the inequality in a single classroom (a small group) versus the whole school. It gives a broader, more nuanced family of measurements.

3. The Challenge: The "Bias" Trap

When you try to estimate inequality from a small sample (like asking 10 people instead of 10,000), your ruler tends to be slightly off. This error is called bias.

The Paper's Achievement: The authors did the heavy math to create a closed-form formula (a precise recipe) to calculate exactly how much this ruler is off.
The Result: They found that while the ruler is slightly crooked for small groups, it becomes perfectly straight as you add more people to the sample. They proved that if you keep adding people, the error eventually disappears completely.

4. The "Common Rate" Rule (and the Exception)

To get their perfect mathematical formulas, the authors had to make a simplifying assumption: they assumed all the "clubs" (the different income groups) shared the same underlying "rate" of how fast money grows or shrinks.

The Analogy: Imagine all three clubs (Low, Middle, High) use the same currency exchange rate. This makes the math solvable.
The Reality Check: The authors knew real life isn't that simple. So, they ran computer simulations (Monte Carlo studies) to see what happens when the clubs have different rates.
The Finding: Even when the "exchange rates" were different (a more realistic scenario), their new estimator still performed very well, staying accurate and beating other methods.

5. The Real-World Test: Italian Income Data

Finally, they didn't just leave it on paper. They took their new ruler and applied it to real data from the Bank of Italy (2008).

What they found: The data was clearly a mix of different groups. A simple single curve didn't fit. But their "Gamma Mixture" model (specifically one with three components) fit the data beautifully.
The Insight: They identified that about 88% of the data was "low income," 11% was "medium-to-high," and a tiny 1% was the "super-rich" tail. Their method successfully measured the inequality within this complex mix.

Summary

In short, this paper:

Invented a new way to measure inequality by looking at groups of people rather than just pairs.
Derived the math to know exactly how accurate that measurement is when dealing with mixed populations.
Proved that the measurement gets better and better as you get more data.
Tested it on real Italian income data and showed it works even when the math gets messy.

They essentially gave economists a more precise, flexible, and reliable tool for measuring the gap between the rich and the poor in complex, mixed societies.

Technical Summary: Income Inequality Estimation with Gamma Mixtures

Problem Statement
Income data typically exhibit right-skewed distributions with long tails, a characteristic often poorly captured by single parametric distributions. Finite mixture models, particularly Gamma mixtures, offer a flexible framework to represent such heterogeneous populations as collections of homogeneous subpopulations. While the classical Gini coefficient ( $m=2$ ) is a standard measure of inequality, the $m$ th Gini index ( $IG_m$ ), introduced by Gavilan-Ruiz et al. (2024), generalizes this concept by utilizing the range of samples of size $m$ . A critical statistical challenge in estimating inequality measures is finite-sample bias. Previous literature has established unbiasedness for the $m$ th Gini index under single Gamma populations (Vila and Saulo, 2025b) and derived bias expressions for the classical Gini coefficient under Gamma mixtures (Vila and Saulo, 2025a). However, a comprehensive analytical treatment of the $m$ th Gini index estimator under finite Gamma mixtures, including its bias, asymptotic properties, and performance under varying scale parameters, was lacking.

Methodology
The paper focuses on a random variable $X$ following a finite mixture of Gamma distributions, denoted $X \sim GM(\theta)$ , where $\theta$ includes mixing proportions, shape parameters, and a common rate parameter $\lambda$ . The authors employ the following methodological steps:

Derivation of Population Index: Using the mixture representation and the multinomial theorem, the authors derive a closed-form expression for the population $m$ th Gini index ( $IG_m$ ). This expression involves integrals of incomplete Gamma functions and multinomial coefficients, characterizing the index as a function of the mixture parameters.
Estimator Analysis: The study analyzes the non-parametric U-statistic estimator ( $\widehat{IG}_m$ ) proposed by Deltas (2003) and extended by Vila and Saulo (2025b). By leveraging the properties of order statistics and the Laplace transform of the Gamma mixture, the authors derive explicit formulas for the expectation and the finite-sample bias of $\widehat{IG}_m$ .
Asymptotic Theory: Assuming a common rate parameter across mixture components, the paper establishes:
- An asymptotic lower bound for the bias using Jensen's inequality.
- Asymptotic unbiasedness, proving the bias vanishes as the sample size $n \to \infty$ .
- Strong consistency and asymptotic normality, utilizing the theory of U-statistics and the multivariate delta method.
Simulation Study: A Monte Carlo study evaluates the estimator's performance (Bias and RMSE) under various scenarios, including different sample sizes ( $n$ ), $m$ values, and shape parameters. The study compares the proposed estimator against a bias-corrected version and a parametric estimator (based on Maximum Likelihood Estimation). Crucially, the simulation also investigates the estimator's behavior when the common rate parameter assumption is violated (i.e., mixtures with different scales).
Empirical Application: The methodology is applied to a 2008 Bank of Italy income dataset. The authors fit Gamma mixtures with 2 and 3 components (both with equal and different scales) to model the income distribution and estimate the $m$ th Gini index.

Key Results

Analytical Expressions: The paper provides the first closed-form expression for the $m$ th Gini index under Gamma mixtures and explicit formulas for the expectation and bias of its U-statistic estimator.
Bias Properties: Theoretical results confirm that the estimator is asymptotically unbiased. The bias is shown to depend on the sample size and the mixture parameters but is independent of the rate parameter due to scale invariance.
Simulation Findings:
- The bias of the non-parametric estimator is generally small and decreases as $n$ increases.
- For small sample sizes, the bias-corrected estimator performs better in terms of bias, while the parametric estimator often yields the lowest Root Mean Square Error (RMSE), despite potentially higher bias.
- The estimator demonstrates robustness even when the underlying data follows a Gamma mixture with different rate parameters, a scenario not covered by the strict theoretical assumptions. In these cases, the estimator maintains low bias and RMSE, though it exhibits a slight negative bias.
Empirical Application: The analysis of the Italian income data suggests that a Gamma mixture with three components (capturing low, medium-high, and top-income segments) provides a superior fit compared to single Gamma or two-component models, as evidenced by QQ-plots and information criteria (BIC). The estimated $m$ th Gini indices vary slightly depending on the model specification (number of components and scale assumptions).

Significance and Claims
The paper claims to extend the theoretical understanding of inequality estimation from single Gamma populations to the more complex and realistic setting of Gamma mixtures. Its primary contribution lies in bridging the gap between the theoretical properties of the $m$ th Gini index and its practical application in heterogeneous populations.

The authors modestly assert that while their rigorous asymptotic results (unbiasedness, consistency, normality) rely on the assumption of a common rate parameter, the Monte Carlo study suggests the proposed methodology remains effective and useful even when this assumption is relaxed. The work provides a flexible framework for measuring income inequality that accounts for population heterogeneity, offering a practical tool for researchers dealing with skewed income data. The paper concludes that future research could extend these results to mixtures with distinct rate parameters and other families of positive distributions.