Bias in Gini coefficient estimation for gamma mixture populations
Roberto Vila and Helton Saulo · 2025
Read Bias in Gini coefficient estimation for gamma mixture populations by Roberto Vila and Helton Saulo (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-00690.txt.
Introduction
Roberto Vila and Helton Saulo study bias in estimating the Gini inequality measure for mixed gamma populations, in a paper from 2025. The Gini index sums up how uneven a distribution is. For a single gamma law, the standard estimator is unbiased, as prior work showed.
For mixtures of gamma laws, they prove the same estimator is biased and derive an exact formula for that bias. The formula grounds a corrected estimator that removes the bias in principle. Monte Carlo simulations then test how the corrected version behaves across sample sizes and mixture shapes. The work matters wherever incomes or sizes mix several gamma groups.
Catalog record
| CREATOR | Roberto Vila and Helton Saulo |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 32 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-00690
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Bias in Gini coefficient estimation for gamma mixture populations Roberto Vila and Helton Saulo Abstract 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. Keywords. Gamma mixture distribution, Gini coefficient estimator, biased estimator. Mathematics Subject Classification (2010). MSC 60E05 \cdot MSC 62Exx \cdot MSC 62Fxx. 1 Introduction The Gini coefficient is a widely used measure of income inequality and dispersion in a population ( Dorfman,, 1979 ) . The degree of inequality within a distribution can be quantified using this method, which has applications in a variety of fields, including economics, finance, ecology, health, and environment, among others; see, for example, Damgaard and Weiner, (2000) , Yao, (1999) , Sun et al., (2010) , and Kharazmi et al., (2023) . According to Baydil et al., (2025) , the Gini coefficient is commonly utilized by the World Bank in order to evaluate the degree of economic disparity that exists across countries. This highlights the practical significance of the Gini coefficient. Mathematically, the Gini coefficient is commonly defined in terms of the expected absolute differences between two independent, identically distributed (i.i.d.) random variables. A widely used estimator for this index is the sample Gini coefficient; see Deltas, (2003) . However, this estimator tends to be biased downward, particularly in small samples and across various distributions, including uniform, log-normal, and exponential models. To address this issue, Deltas, (2003) introduced the upward-adjusted sample Gini coefficient, which was shown to be unbiased when the population followed an exponential distribution. Recently, Baydil et al., (2025) extended the unbiasedness of the upward-adjusted sample Gini coefficient when the population follows a gamma distribution. The gamma distribution is particularly appealing in economic modeling due to its flexibility in capturing mid-range income distributions, as opposed to the log-normal model, which is often used for high-income distributions; see Salem and Mount, (1974) . McDonald and Jensen, (1979) derived the population Gini coefficient for a gamma-distributed population and explored its estimation via maximum likelihood and moment-based methods. Nevertheless, a single gamma distribution for modeling income may not be adequate as real-world income data is often heterogeneous, consisting of multiple economic classes (e.g., low-, middle-, and high-income earners). In contrast, a gamma mixture model can be a more flexible parametric model for an income distribution as studied by Chotikapanich and Griffiths, (2008) . In fact, mixtures offer the benefit of a flexible functional structure while maintaining the convenience of parametric models that facilitate statistical inference; see Chotikapanich and Griffiths, (2008) . In this paper, we examine the estimation of the Gini coefficient in gamma mixture populations, a more general class of distributions that accounts for heterogeneity in income data. While the sample Gini coefficient is known to be an unbiased estimator for the population Gini coefficient in single gamma distributions ( Baydil et al.,, 2025 ) , we show that this property does not hold for gamma mixture models, leading to a systematic bias in estimation. Then, we derive an explicit expression for the bias and propose an unbiased Gini estimator when the population is gamma mixture distributed. The rest of this paper unfolds as follows. In Section 2 , we present the theoretical foundations and key definitions. In Section 3 , we derive a closed-form expression for the expectation of the sample Gini coefficient estimator. In Section 4 , we demonstrate the bias introduced when estimating the Gini coefficient from gamma mixture distributions, providing both theoretical and empirical evidence. In Section 5 , we explore an Illustrative Monte Carlo simulation study. Finally, in Section 6 , we provide some concluding remarks. 2 Preliminary results and some definitions This section provides the theoretical foundation for our analysis, introducing preliminary results and definitions. A key result, Proposition 2.4 , presents an explicit formula for the Gini coefficient, a widely used statistical measure of income inequality ( Gini,, 1936 ) . The theoretical results presented in this section, pertaining to (almost surely) positive random variables, exhibit universal validity and applicability, transcending specific distributional forms and accommodating real-valued random variables with diverse support structures. Lemma 2.1 . Let X_{1} and X_{2} be two independent copies of a (almost surely) positive random variable X with finite integral and common cumulative distribution function F and let g be a positive real-valued integrable function of two (positive) variables. Then, the following identity holds: \displaystyle\mathbb{E}[|X_{1}-X_{2}|g(X_{1},X_{2})]=2\mathbb{E}(X)\left\{\mathbb{E}\left[g(X^{*},X)\mathds{1}_{\{X<X^{*}\}}\right]-\mathbb{E}\left[g(X,X^{*})\mathds{1}_{\{X^{*}<X\}}\right]\right\}, where \mathds{1}_{A} is the indicator function of an event A , X^{*} is independent of X and has length-biased distribution, that is, its cumulative distribution function is given by the following Lebesgue-Stieltjes integral: \displaystyle F_{X^{*}}(u)={1\over\mathbb{E}(X)}\,{\int_{0}^{u}t{\rm d}F(t)},\quad u>0. (1) Proof. From independence of X_{1} and X_{2} , it follows that \displaystyle\mathbb{E}[|X_{1}- \displaystyle X_{2}|g(X_{1},X_{2})] \displaystyle=\int_{0}^{\infty}\left[\int_{u}^{\infty}(v-u)g(u,v){\rm d}F(v)\right]{\rm d}F(u)+\int_{0}^{\infty}\left[\int_{0}^{u}(u-v)g(u,v){\rm d}F(v)\right]{\rm d}F(u) \displaystyle=2\int_{0}^{\infty}u\left[\int_{0}^{u}g(u,v){\rm d}F(v)\right]{\rm d}F(u)-2\int_{0}^{\infty}\left[\int_{0}^{u}vg(u,v){\rm d}F(v)\right]{\rm d}F(u), (2) where in the last identity we use the fact that X_{1} and X_{2} are identically distributed. By using the definition ( 1 ) of F_{X^{*}} , the expression in ( 2 ) can be written as \displaystyle 2\mathbb{E}(X)\int_{0}^{\infty}\left[\int_{0}^{u}g(u,v){\rm d}F(v)\right]{\rm d}F_{X^{*}}(u)-2\mathbb{E}(X)\int_{0}^{\infty}\left[\int_{0}^{u}g(u,v){\rm d}F_{X^{*}}(v)\right]{\rm d}F(u). Hence the proof readily follows. ∎ Proposition 2.2 . Under the conditions of Lemma 2.1 , if g is symmetric, that is, g(u,v)=g(v,u) , u,v>0 , then \displaystyle\mathbb{E}[|X_{1}-X_{2}|g(X_{1},X_{2})]=2\mathbb{E}(X)\left\{2\mathbb{E}\left[g(X^{*},X)\mathds{1}_{\{X<X^{*}\}}\right]-\mathbb{E}\left[g(X^{*},X)\right]+\mathbb{E}\left[g(X^{*},X)\mathds{1}_{\{X=X^{*}\}}\right]\right\}. Proof. The proof is immediate, therefore omitted. ∎ Proposition 2.3 . Under the conditions of Proposition 2.2 , we have \displaystyle\mathbb{E}(|X_{1}-X_{2}|)=2\mathbb{E}(X)\left[2\mathbb{P}(X<X^{*})-1+\mathbb{P}(X=X^{*})\right]. (3) Equivalently, \displaystyle\mathbb{E}(|X_{1}-X_{2}|)=2\mathbb{E}(X)\left[2F_{X\over X+X^{*}}\left({1\over 2}\right)-1+\mathbb{P}(X=X^{*})\right], (4) with F_{X/(X+X^{*})} being the cumulative distribution function of the independent ratio {X/(X+X^{*})} . Proof. By taking, in Lemma 2.1 , g as the unitary constant function, we obtain ( 3 ). Identity in ( 4 ) follows from the equality of events \{X<X^{*}\} and \{2X<X+X^{*}\} . ∎ The Gini coefficient is defined using the Lorenz curve, which plots cumulative income share against population percentage. The line of perfect equality is a 45-degree line. The Gini coefficient is calculated as the ratio of the area between the Lorenz curve and the line of equality to the total area under the line. To derive explicit expressions, we use the standard definition based on the mean difference. Definition 2.1 . The Gini coefficient ( Gini,, 1936 ) of a random variable X with finite mean \mathbb{E}(X) is defined as \displaystyle G={1\over 2}\,{\mathbb{E}(|X_{1}-X_{2}|)\over\mathbb{E}(X)}, (5) where X_{1} and X_{2} are independent copies of X . Proposition 2.4 . Under the conditions of Proposition 2.3 , the Gini coefficient for X is given by \displaystyle G=2F_{X\over X+X^{*}}\left({1\over 2}\right)-1+\mathbb{P}(X=X^{*}), with F_{X/(X+X^{*})} being the cumulative distribution function of the independent ratio {X/(X+X^{*})} and X^{*} distributed according to ( 1 ). We proceed by defining the gamma mixture distribution for the case of a constant rate parameter, \lambda>0 . While our results can be extended to the more general case of non-constant rate parameters, our primary goal is to investigate the bias in the estimation of the Gini coefficient for a gamma mixture distribution, and we demonstrate