TLIC: An R package for the LIC for T distribution regression analysis

Abstract

This paper introduces the TLIC R package, a novel framework that integrates the T-distribution with the Length and Information Criterion (LIC) to address optimal subset selection in regression models with T-distributed errors. Traditional subset selection methods, such as beta_AD, beta_cor, and LICnew, assume normality of errors, which may lead to biased results when dealing with heavy-tailed or skewed distributions. Through extensive simulation experiments, we demonstrate that TLIC outperforms these methods in terms of stability and sensitivity, especially under non-normal error distributions. An R package implementing the TLIC method is also developed, providing a practical tool for researchers to conduct subset selection with T-distributed errors. Our findings highlight TLIC's potential to improve subset selection accuracy in real-world applications where error distributions deviate from normality.