Evaluating the Efficiency of the Jackknife Kibria-Lukman M-Estimator: A Simulation-Based Comparative Analysis
Ayanlowo, E.A
Department of Basic Sciences, Babcock University, Ilishan-Remo, Ogun State, Nigeria.
Oladapo, D.I
Department of Mathematical Sciences, Adeleke University, Ede, Osun State, Nigeria.
Phillips, S.A.
Federal School of Statistics, Ibadan, Oyo State, Nigeria.
Obadina, G.O *
Olabisi Onabanjo University, Ago-Iwoye, Ogun State, Nigeria.
*Author to whom correspondence should be addressed.
Abstract
Although linear regression is frequently used in predictive analysis, the Ordinary Least Squares (OLS) estimator's accuracy is decreased by multicollinearity and outliers. In order to offer a reliable substitute, this study suggests the Jackknife Kibria-Lukman (JKL) M-Estimator, which combines Ridge shrinkage, Jackknife resampling, and M-estimation. In extreme multicollinearity settings with outliers, the JKL M-Estimator reduced MSEs by up to 50% when compared to OLS and 30% when compared to Ridge using Monte Carlo simulations. Furthermore, across estimators, the JKL M-Estimator consistently offered the lowest variation. The JKL M-Estimator reduced the average coefficient variance by 44% when compared to OLS and 25% when compared to Ridge when used to real-world economic data, demonstrating improved resistance to outliers and multicollinearity. These findings confirm that the JKL M-Estimator is a very accurate and stable estimator for real-world regression situations that defy conventional wisdom.
Keywords: Jackknife resampling, multicollinearity, M-estimation, robust regression, outliers