Scholarlink

<p>The ridge regression estimator (RRE) is widely used as an improved estimator for estimating regression parameters in multicollinear linear regression models. However, an argument exists that in the presence of outliers, the dataset may adversely affect these improved estimators (<i>Australian J. Stat.</i> <b>33</b>(3), 319–333, 1991). This paper proposes several improved estimators and their robust versions to address the multicollinearity problem, regardless of the presence of outliers. To evaluate these estimators, both Monte Carlo simulations and two real-life datasets were utilized, considering various outlier scenarios. We consider the smaller Mean Squared Error (MSE) value as a performance criterion. The simulation study demonstrates that when the dataset contains no outliers, the improved estimators generally outperform most of the robust versions, except for (RRE and RKL). However, in the presence of outliers, all the robust versions of these improved estimators perform better than all conventional improved estimators.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Performance of Some Improved Estimators and their Robust Versions in Presence of Multicollinearity and Outliers

  • 期刊论文

  • 发表日期：

    2025年2月4日

查看全文
  • Nusrat Yasmin,
  • B. M. Golam Kibria

摘要

The ridge regression estimator (RRE) is widely used as an improved estimator for estimating regression parameters in multicollinear linear regression models. However, an argument exists that in the presence of outliers, the dataset may adversely affect these improved estimators (Australian J. Stat. 33(3), 319–333, 1991). This paper proposes several improved estimators and their robust versions to address the multicollinearity problem, regardless of the presence of outliers. To evaluate these estimators, both Monte Carlo simulations and two real-life datasets were utilized, considering various outlier scenarios. We consider the smaller Mean Squared Error (MSE) value as a performance criterion. The simulation study demonstrates that when the dataset contains no outliers, the improved estimators generally outperform most of the robust versions, except for (RRE and RKL). However, in the presence of outliers, all the robust versions of these improved estimators perform better than all conventional improved estimators.