ABSTRACT This paper examines the adequacy of various linear and generalized linear models in analyzing income distribution, specifically focusing on the Gini index, which is constrained within the (0,1) interval. We compare traditional linear models, including pooled, fixed effects, and random effects approaches, with beta regression models that are better suited for data bounded between 0 and 1. Hypothesis tests—Chow, Breusch‐Pagan, and Hausman—are employed to determine the appropriateness of these models, revealing that fixed effects and random effects models are preferable over pooled data models. While beta regression models offer a theoretically sound approach due to their alignment with the bounded nature of the Gini index, empirical results show that the random effects models (both linear and beta) yield similar numerical outcomes for the covariates. This suggests that linear models, despite their simplicity, provide a comparably effective description of the data. Additionally, the analysis of R 2 and pseudo‐ R 2 metrics highlights the superior fit of beta regression models, suggesting their greater effectiveness in capturing the nuances of income distribution. The paper's novel contribution lies in its direct empirical comparison of beta regression and traditional panel data models in the context of income distribution, an area where such comparative analysis has been limited. Policy implications highlight that while inflation control and GDP growth are necessary, they are not sufficient to reduce inequality, calling for long‐term, multifaceted interventions. Future research should expand the dataset to include more variables and countries to validate these findings.
Diniz et al. (2026) studied this question.