Evaluating the Robustness of Some Two-Sample inferential Statistics in the Presence of Mixture Distributions: A Simulation study

Le résumé fourni par la source

This study investigates the robustness of two-sample inferential statistics when datasets are derived from mixture distributions, where traditional methods like the t-test may fail due to violated assumptions. Using R software, random variables from Standard Normal, Gamma, and Exponential distributions were generated and analyzed using four inferential tests: Rank Transformation t-test (Rt), Wilcoxon Sum Rank Test (WSD and its Asymptotic version WSA), and Trimmed t-test (Tt-test). Robustness was evaluated based on Type I error rates across varying levels of multicollinearity and sample sizes (n=10, 20, 30, 40, 50, 60, 70, 80 and100). A test was deemed robust if it maintained acceptable error rates (α=0.1, 0.05, and 0.01) and demonstrated consistency across multicollinearity levels and sample sizes. At α=0.1, the WSD and Tt-test exhibited the highest robustness. At α=0.05, the Tt-test was the most robust, while at α=0.01, both the Tt-test and WSD were robust, with the Tt-test slightly outperforming. Overall, the Tt-test and WSD consistently demonstrated robustness across all significance levels, suggesting they are reliable alternatives for two-sample problems involving mixture distributions. These findings underscore the importance of selecting robust statistical methods to ensure accurate inferences in complex data scenarios.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.